Water quality monitoring is critical for sustainable aquaculture, as it directly affects fish health, growth, and productivity. Traditional monitoring methods, reliant on manual sampling and laboratory testing, are labor-intensive, time-consuming, and subjective. Although numerous artificial intelligence (AI) techniques have been proposed to forecast individual water quality parameters, the challenge of assessing overall water quality by combining multiple sensor inputs into a meaningful quality category remains underexplored. To address this gap, we propose a two-stage approach consisting of two complementary systems: a Mamdani–Assilian fuzzy inference system (MAFIS) for water quality labeling and a Takagi–Sugeno–Kang compact adaptive neuro-fuzzy system (TSK-CANFS) for predictive modeling. MAFIS utilizes expert-defined membership functions to label water quality based on individual parameter values, effectively generating labeled datasets for training. TSK-CANFS, on the other hand, leverages a novel rule generation mechanism and the sliding window approaches to construct a reduced fuzzy rule base for efficient and accurate water quality prediction. Experiments on real-world water quality datasets demonstrate that the proposed MAFIS and TSK-CANFS achieve competitive performance compared to state-of-the-art methods. Combining data-driven learning with expert-defined reasoning, our approach enhances interpretability, scalability, and accuracy, offering a practical solution for sustainable aquaculture management.

The diagnosis of high myopia using fundus images is essential for visual health. Existing deep learning-based methods rely on large-scale labeled data, but annotated data for high myopia fundus images remains scarce. To address this issue, we mimic the ability of ophthalmologists to diagnose a new disease with only a small number of samples. In this regard, we propose a meta-learning network guided by domain knowledge from fundus images for diagnosing high myopia. The model consists of three modules: First, the image reconstruction module builds an autoencoder (comprising an encoder and a decoder) that takes the input raw image and outputs the reconstructed image. Next, the fundus image domain knowledge learning module constructs a Siamese network to learn the similarity between the original and reconstructed fundus images. This similarity is used as a loss function to guide the encoder in effectively learning the domain features of fundus images. Finally, in the domain-knowledge-guided meta-learning module, the encoder’s initialization parameters (obtained from the first two modules) are further optimized using the OCMAMAL architecture, resulting in more optimal encoder parameters. This optimization helps achieve superior recognition performance with only a small amount of data for new tasks. Using a clinically real high myopia fundus image dataset, our method achieved F1 scores of 83.4%, 87.9%, and 89.2% under 5-shot, 10-shot, and 20-shot conditions, respectively, demonstrating the effectiveness of the proposed method.

Diabetic retinopathy (DR) is the leading cause of blindness in diabetic patients, which necessitates precise grading of retinal lesions for early diagnosis. Existing DR grading methods typically employ image enhancement techniques to improve the quality of fundus images. However, due to variations in imaging devices and differences in the proficiency of medical practitioners, the quality of images often exhibits significant heterogeneity. Uniform enhancement across all fundus images may inadvertently amplify noise artifacts, particularly in high-quality images. Moreover, since diabetic lesions in fundus images are often small, reliance solely on global image features makes it difficult to fully capture fine-grained lesion features. To address these challenges, this paper proposes a dual-stream deep learning model that integrates quality-aware dynamic enhancement and a multi-instance multi-scale vision transformer. First, An image quality assessment-based selective enhancement strategy was implemented, wherein only low-quality fundus images underwent enhancement processing. Then, a dual-branch processing architecture is designed to differentially handle enhanced and non-enhanced images. Experimental results on real-world datasets demonstrate the effectiveness of the proposed method.

Fusing medical images from diverse modalities like MRI, PET, and SPECT helps improve diagnostic precision by combining their complementary features. However, existing deep learning-based fusion methods often suffer from limited detail preservation and inefficient attention modeling across spatial and channel dimensions. This study introduces an innovative framework called hierarchical attention fusion network (HAF-Net) for robust and high-quality medical image fusion. The proposed model incorporates a hierarchical feature aggregation (HFA) module to extract scale-adaptive features, and a residual attention convolution (RAC) block to enhance fine-grained details using gradient-aware spatial and frequency-domain information. Furthermore, a multispectral frequency-aware channel attention (MFCA) mechanism is introduced to capture discriminative features across multiple frequency bands, and a cross-interaction attention module (CIAM) is designed to jointly model spatial-channel relationships. An adaptive fusion weighting (AFW) strategy is employed to dynamically combine multi-scale features based on their contextual relevance. Extensive experiments on standard PET/MRI and SPECT/MRI datasets demonstrate that HAF-Net achieves superior performance compared to state-of-the-art fusion methods. The results validate the effectiveness of the proposed modules in preserving structural integrity and enhancing detail in fused medical images.

Accidents involving escalators in mass rapid transit (MRT) systems pose a serious risk to public safety, often resulting from clothing or footwear getting caught, or large items toppling during movement. Despite the availability of passive warnings, such as signage and audio announcements, these methods often go unnoticed by commuters and lack the ability to adapt to real-time risks. Existing computer vision solutions are either too computationally intensive for deployment on edge devices or lack sufficient accuracy for practical use. To address these challenges, this study proposes a real-time, lightweight object detection system using a pruned YOLOv7-Tiny model, optimized for deployment on the NVIDIA Jetson Nano edge computing platform. The system is designed to identify safety-critical items, such as general footwear, high heels, long skirts, suitcases, strollers, and shopping trolleys, in real-time. Upon detection, it issues visual and auditory alerts, and in cases involving large items, sends email notifications to station personnel. Model pruning significantly reduces computational overhead while maintaining high accuracy. Experimental results demonstrate that the system achieves a mean average precision (mAP) of 94.69%, outperforming conventional detection models while maintaining real-time performance. These results highlight the system’s potential for enhancing passenger safety and operational efficiency in resource-constrained public transit environments.

Bird excrement deposited on solar panels can lead to hotspots, significantly reducing the efficiency of solar power plants. This article presents a novel solution to this problem leveraging unmanned aerial vehicle (UAV) systems for the automated geolocation and removal of bird excrement across large-scale solar power facilities. First, a UAV executes a predefined flight path to capture sequential aerial images of the plant. These images are subsequently stitched to produce a high-definition orthomosaic of the entire facility. An advanced detection framework based on YOLOv7, enhanced with an attention module, is employed to accurately detect bird excrement by reducing background noise and highlighting key features. An additional prediction head is integrated to improve detection of smaller bird excrements. To compute precise geolocation of the detected excrement, the midpoint pixel coordinates of the excrement along with the azimuth angle and actual ground distance (AGD) relative to a ground control point (GCP) is used. This article further proposes a cleaning technique that employs a traveling salesman problem (TSP) approximation algorithm to efficiently optimize flight path of the cleaning UAV. Experimental results indicate the system achieves an average detection precision (AP) of 93.91% and GPS coordinate accuracy with an average error of 0.149 m, demonstrating the efficacy of the proposed method in both geolocation and removal of bird excrement from solar panels.

Feature selection holds significant importance in knowledge mining as it plays a pivotal role in selecting and preserving the most informative features within a dataset while discarding irrelevant, redundant, or noisy attributes. This process contributes to enhancing model performance, reducing computational complexity, and refining interpretability, thus facilitating more accurate and efficient data analysis. In high-dimensional datasets, the necessity for feature selection becomes more pronounced due to the heightened risk of encountering the curse of dimensionality. Therefore, this study proposes a weighted fuzzy rough quickreduct (FRQR) feature selection approach employing feature weights to handle the equal situation problem inherent in FRQR. The proposed method is evaluated on ten publicly available datasets with feature sizes ranging from 2000 to 15154. The selected features are used to train and test random forest candidate models whose estimates are then combined according to the posterior probabilities by Bayesian Model Averaging (BMA). The performance of the model on the selected features is evaluated on four performance metrics. The essentiality of the selected features is further determined by comparing the model classification performance achieved on the non-selected features and all the dataset features. The results indicate competitiveness in the performance metric values achieved on selected features over the other two feature categories affirming the efficacy of the proposed method.

In the era of rapidly advancing technology, food classification has emerged as a pivotal application across various domains including health monitoring, dietary assessment, and culinary innovation. However, efficiently categorizing food items remains a challenge, particularly in scenarios with limited labeled data. This paper introduces a novel approach for few-shot food classification using Prototypical Networks with ResNet-50 and an attention mechanism as embedding network. Leveraging the inherent capability of Prototypical Networks to learn from scarce examples, our method demonstrates exceptional adaptability and accuracy in classifying food items. Through extensive experimentation on the Food-101 dataset, employing various CNN architectures, our findings underscore the effectiveness of our approach. In particular, ResNet-50 integrated with the attention mechanism surpasses other architectures, achieving superior classification accuracies of 91.5% and 95.2% for 1-shot and 5-shot learning scenarios, respectively. This integrated approach showcases the potential of Prototypical Networks in addressing the challenges of limited labeled data in food classification tasks, marking a significant advancement in the field.

Recently, the coronavirus disease of 2019 (COVID-19), as named by the World Health Organization (WHO), has spread to over 200 countries. The WHO has declared this disease as a worldwide public health emergency. One of the most difficult tasks in combating this epidemic is to identify and segregate the afflicted people. The reverse transcription-polymerase chain reaction test (RT-PCR) is the most common pathology test used to diagnose this infection. Studies show that the RT-PCR test has a low-positive rate and sometimes becomes ineffective in diagnosing infection. In some cases, computed tomography (CT) scans reveal acute pneumonia and pulmonary anomalies. Therefore, CT scans are used together with RT-PCR tests to confirm infected people. Existing artificial intelligence and machine learning techniques require a large number of CT scans for training, which is a time-consuming process. Visual inspection shows that most CT scans of COVID-19 cases have broken, blurred, and ambiguous edges for infectious areas. Another major issue with these images is the heterogeneous intensity of the pixels, high noise, and low resolution. As a result of all these issues, the problem of effective edges/boundaries of various areas of CT scans of COVID-19 cases cannot be resolved by the current edge detection approach. Indeed, improper selection of edges can lead to an incorrect diagnosis of diseases through CT scans of COVID-19 cases. Therefore, there is an urgent need for a diagnostic method in addition to the RT-PCR test that can extract useful information from the minimum number of chest CT scans of suspected COVID-19 cases. This study introduces a new ambiguous edge detection method (AEDM) for identifying the edges/boundaries of different regions in CT scans of COVID-19 cases. The proposed AEDM is developed on the basis of ambiguous set (AS) theory, which is highly efficient in processing ambiguous pixel information. For simulation purposes, various CT scans of COVID-19 cases are classified into three different categories: 1) low infection (LI); 2) moderate infection (MI); and 3) severe infection (SI). Empirical analysis shows that the proposed AEDM can effectively highlight the edges in CT scans of three different categories in comparison with other well-known edge detection methods.

Conventional soft clustering algorithms perform well on linearly distributed features, but their performance degrades on nonlinearly distributed features in high-dimensional space. In this study, a novel soft clustering algorithm, the ambiguous kernel distance clustering (AKDC) algorithm, is presented. This algorithm is developed by applying ambiguous set theory and the Gaussian kernel function. The ambiguous set theory defines the ambiguities inherent in each feature with four membership values: 1) true; 2) false; 3) true-ambiguous; and 4) false-ambiguous. The degree of membership values here forms a low-dimensional feature space that is not linearly distributed. Therefore, these nonlinearly distributed membership values are mapped into a high-dimensional feature space using the Gaussian kernel function. This study focuses on performing cluster analysis of computerized tomography scans of COVID-19 (CTSC-19) cases using AKDC. COVID-19, recognized as one of the most life-threatening diseases of this century, is highly contagious, and early diagnosis may prevent one-to-one transmission. Extensive empirical studies have been conducted with different types of CTSC-19 to demonstrate its effectiveness against existing kernel-based clustering and nonkernel-based clustering algorithms, namely mercer kernel fuzzy c-mean (MKFCM), kernel generalized FCM (KGFCM), kernel intuitionistic fuzzy entropy c-means (KIFECMs), morphological reconstruction and membership filtering clustering (FRFCM), and intuitionistic FCM based on membership information transferring and similarity measurements (IFCM-MS). The effectiveness of the proposed algorithm compared to the existing algorithms is evaluated using standard statistical metrics, such as dice index (DI), Jaccard index (JI), structural similarity index (SI), and correlation coefficient (CC). The empirical results show that AKDC is more effective than existing algorithms based on DI, JI, SI, and CC.

The accumulation of bird droppings on photovoltaic (PV) farms reduces power generation efficiency and necessitates manual cleaning on a regular basis, which is a challenge in large power plants. To solve this problem, this paper proposes an automatic Unmanned Aerial Vehicle (UAV) based bird droppings detection, localization, and cleaning method on large PV power plant. An automated flight route is first created, and use an UAV to fly over the solar farm to capture images of the solar panels. The captured images are then stitched together to create a high-resolution orthomosaic image of the solar farm, which enables to precisely locate the bird droppings on the solar farm. An improved YOLOv7-based model is proposed to detect the bird droppings because they are quite small in comparison to the stitched image. Then, using the ground sample distance, we calculate the distance between each of the bird droppings and the drone's takeoff point, which is used to clean the bird droppings from the solar panel. Last, the proposed model is verified by high-resolution orthomosaic images and the experimental outcomes unequivocally show that it is successful for detecting and cleaning of bird droppings on PV farms.

Distinct defects will inevitability incur due to multiple layers of production are required to manufacture TFT-LCD panels. Localization and segmentation of defects are vital in monitoring the panels to improve the yields. But accurate segmentation of defect areas is challenging due to the complex/diverse defects, illumination artifacts, the similarity of the defective pixels with the neighboring pixels, and excessive overall colors in the image. Furthermore, defects have distinct shapes, types, sizes, and locations. This study proposed a singular value decomposition Kernelized Neutrosophic entropy (SVDKNE) method to resolve these challenges that can enhance the inhomogeneous defect images adaptively. Finally, the comparative analysis on a dataset with 309 images validates that the SVDKNE method outperforms other five methods in locating and segmenting defects with higher values of average Jaccard similarity of 0.93, structural similarity of 0.99, and peak signal-to-noise ratio of 38.78.

This research presented a new segmentation method based on fuzzy set, entropy and image fusion to analyze brain tumors from magnetic resonance imaging (MRI). Using fuzzy set, one can tackle the problem of uncertainty representation in gray levels of MRIs during the segmentation process. This uncertainty in their gray levels occurred due to poor illumination of images. To resolve this issue, this study focused on fuzzification of gray levels and assignment of membership degrees based on membership functions. Each fuzzified gray level value was quantified using entropy. The proposed method generated multiple thresholds based on maximum entropy values of gray levels. These thresholds generated multiple segmented images with different features. Finally, image fusion operation was performed on multiple segmented images to highlight all the critical features of brain tumors. Fusion images were compared with the segmented images obtained from four additional methods, the multilevel threshold method, adaptive threshold method, K-means clustering algorithm and fuzzy c-means algorithm. The performance evaluation metrics indicated the effectiveness of the proposed method over these existing methods.

This article presented a new set theory to deal with ambiguousness, which was entitled as an “Ambiguous Set Theory”. The proposed ambiguous set theory can represent any feature into four degrees of memberships, viz., true, false, ambiguous-true and ambiguous-false. This kind of representation provides granular visualization of features, and helps to model uncertainties very effectively. In this article, initially we discussed the motivation to introduce the theory of ambiguous set. Then, we proposed methodology of ambiguous set by: 1) defining it in a precise way, 2) presenting a mathematical representation for the set, and 3) giving various mathematical definitions for the set. Applications of the proposed ambiguous set were demonstrated in human brain MRI segmentation. Various comparison results demonstrated the effectiveness of the theory over existing well-known approaches of the image segmentation.

A method is presented for scene detection and estimation using high-resolution imagery acquired through autonomous drone navigation aided with landmark detection and recognition. The proposed system comprises a drone platform that facilitates efficient autonomous flight; it can capture images and provide real-time video streaming of the ground cover using a camera equipped with a 14-megapixel CMOS sensor and a fish-eye lens. In addition, landmark detection and recognition was performed by applying the histogram of oriented gradients and linear support vector machine methods on each frame of the video stream. The high spatial resolution of the acquired drone images makes the detection and interpretation of environments less complicated. First, through image processing, orthomosaic images and 3-D environment reconstruction (point clouds) of the scene are generated from a set of drone images by using an automatic photogrammetric technique called “structure from motion.” Subsequently, an unsupervised classification method is used to detect and differentiate environmental classes (scene interpretation) in the target or investigated area by using the high-resolution images. Finally, the results of the proposed method are evaluated by comparing them against ground-truth points.

Social network websites have become important marketing platforms for studying business models. When impressive advertisements are posted on the platforms, social network users like to comment or post their experiences as part of feedback about the products. Those interesting opinions will exert influences on other users buying decisions but some may remain commented. This invokes peoples interests to dig out interesting relationship between sentiment of social network users and the volume of product sales. Then, they can apply the discovered patterns of sentiment and sales to predict users buying behaviors. This study proposes a framework based on data mining method to find interesting patterns of sentiment and sales. The proposed model starts by defining sentiment topics with their corresponding terms and then follows by a fuzzy model to infer the sentiment scores for user opinions. Each transaction in the database is transformed to attach with public sentiment scores, influential users sentiment scores and volume of product sales. To better obtain the relationship among public sentiment, users sentiment and volume of product sales, a mining method of inter-transaction association rules is considered to extract the interesting patterns of sentiment and sales. Two case studies are given to verify the effectiveness of the proposed method.

The current populace of the elderly is apparently abandoned by the younger generations due to their individual circumstances. To heighten the vitality and strengthen the fitness of elders, assisting a home care system can be an admittance that provides comprehensive nursing and monitoring them in the regular interim. To deliver an interactive service supervision platform to the elders a smart environment of various sensors are clubbed together to establish an intuitive platform that can control the home appliances and gadgets within the living space of elders. The proposed system used voice and gesture (MPU6050 accelerometer) to control the home appliances like turning on/off the light, closing/opening of curtains, TV, and fan or AC within the living spaces. The system also monitors the real-time activity like heart rate and body temperature for the elderly citizens. In the case of emergency, for instance, anomalous behaviors like heart stroke occurs, the proposed system set-up triggers an alarm and the emergency bulb will be strikes "on" to alert their kin. This smart environment can set the temperature and help control the living parameters based on the users' comfort and their health conditions. The whole design is to provide modest support systems for the elder to live healthily and safely in an independent living environment.

The paper proposes a data mining method to find the relationships between fan page users' sentiment and customers' purchase behavior in order to predict those customers' purchase intention in the future. The business companies create their own fan pages and post advertisements to prompt their products. The fan page users always post their opinions on the wall to tell the feelings about products. Since all those opinions will be spread to every corner in the social network, some marketing managers would wonder whether those opinions help or harm the products sale. If we can measure the sentiment of users' opinion and then find the relationships between users' sentiment and products sales volume, the discovered sentiment-sales patterns can be used to predict their customers' future purchase intention. In this study, a framework based on fuzzy set model and association rule mining is proposed to find sentiment-sales patterns. First, the specific sentiment topics with their related terms need to be defined. Then the fuzzy membership functions for the terms are used to evaluate fan page users' sentiment score. Assuming the products sales volume information is released from the company, and we can use inter-transaction association rule mining method to find the sentiment-sales patterns which reveal the relationships between fan page users' sentiment and customers' purchase behavior. A theoretical experiment is given to illustrate how the proposed framework works.

This paper presents a method that can detect driver's drowsiness by using the wavelet packet transform (WPT) and functional link-based fuzzy neural network (FLFNN) models. Drowsy drivers have been reported to be vulnerable to car accidents. Early detection of drowsiness can help alert drivers or passengers to provide a safety drive on the road. For those old models or cars without equipped with advanced high technologies, there is a dire need to install sensor devices that can effectively detect drowsy status of drivers at an early stage. Photoplethysmography (PPG) is a non-invasive optical technique that measures relative blood volume changes in the blood vessels and has been universally used for research and physiological study. We develop such PPG sensor devices to be installed on the steering wheel to detect the physiological conditions (such as normal to drowsy) by using parameters extracted from the heart rate variability (HRV) obtained from PPG signal calculation. Experimental results revealed that the proposed model is effective in assessing the drowsy levels of drivers.

Klout, a famous App, could measure people's social network influence power. Klout score is measured according to the data from past 90 days and an individual who has high Klout score is thought as having high social influence power. Lots of businesses or organizations like to hire high Klout score people to help them to diffuse their brand images. However, Klout score cannot tell us who has high influence power in a specific short time period. For example, it is possible that some of the users might always have high influence power on Monday or on Monday morning. These time-dependent influential users probably have low Klout scores in average but have high influence power in some specific time periods. Businesses should not just know who are the high Klout score users but also they should identify who are the time-dependent influential users because all of them may have some sort of power to influence other users' buying decisions. In this study, a framework based on frequent pattern mining is proposed to find the time-dependent influential users. First of all, the framework will divide a predefined long time period into successive short time segments and then influential transactions that contain Facebook fans' influence power data will be defined in each time segment. From the frequent patterns, the proper time for time-dependent influence users to spread information can be found. A theoretical experiment is given to verify the effectiveness of the proposed framework.

Equirectangular panoramas are popular tools for achieving 360 degree immersed viewing experiences. A panorama captures a scene from one point and panoramic viewers allow the user to control the viewing direction, but the viewer is not allowed to move around. This study proposes a strategy for transforming equirectangular panoramic images with the effect of moving freely in three dimensions. The strategy assumes that the panorama is an enclosed space comprising a flat ground and flat vertical walls. A equirectuangular Hough transform is proposed for detecting the boundaries of the respective planes. The panoramic image is then decomposed into the respective planes, the viewing point is translated and a new panoramic image based on the new viewing position is composed. Preliminary proof of concept test shows that the strategy allows free translation within simple panoramic images.

Many skin detection approaches have been proposed in the image analysis literature. Some are simple and static; the others are dynamic and rely on complex machine learning algorithms and training data. Generally the simple approaches are preferred. We hypothesize that the developers' choice for the simple approaches are due to the reasonable quality of results and ease of implementation, since the results of more sophisticated results are not readily available. This paper explores the skin color of a large number of hand samples using color space visualization. The results suggest that a static method may suffice for many applications, but that a small set of rules is not enough to capture the details of skin. Moreover, the results suggest that successful skin detection does not depend on the color space used as there are no apparent advantages of using a perceptual uniform color space such as CIElab. A skin detection approach based on RGB-color table-lookup is proposed that is able to capture the complex skin color cluster shape. The method is practical and simple to implement with minimal computational cost. The lookup table is released into the public domain.

Finger lengths have fascinated scientific researchers for a long time and people hardly notice different underpinnings of the relation between the shape and size of our body to the scientific meanings. The idea behind the shape of the palm length of the fingers indicates something profound about people sexual proclivities and vulnerability to certain diseases. In order to investigate and intervene in this health concern we measure the digit ratio that is the ratio of the lengths of different digits. The index finger (2D) and ring finger (4D) for both the hands are typically measured from the midpoint of the bottom crease where the finger joins the hand to the tip of the finger. This study uses smart phone camera to take the hand image and then measure relative lengths of the second (2D) and the fourth (4D) fingers. Their ratio is calculated by dividing the length of the index finger by the length of the ring finger of the hand. This finger reading foretells information of an individual's psychology, motivations, health and social behavior such as homosexuality, cancers, musical ability, aggressive personality, passive personality traits, development disorder such as Dyslexia (which can be termed as literacy deficiencies), athletic ability and dexterity innate ability in key cognitive areas.

Social networks have become an important marketing tools for business to build brand pages to prompt their new products. Fans' user-experience diffusion results in great marketing power that people never seen. Typically, some of the fans in group are influence users. They are market movers which mean they can influence others buying decisions. Businesses can affect online influence users by giving them extra benefits to turn them into spokesmen. However, who is the influential user? What period of time is appropriate for information to spread? In this study, a framework based on frequent pattern mining is proposed to find the influence users as well as the proper time to spread information. The one day 24-hour period can be divided into successive time segments. An influence transaction that contains fans' influence power will be defined in each time segment. After transactions being collected several days, the frequent patterns can be found to deduce the proper time for influence users to spread information. The theoretical experiment is given to show how the proposed framework works.

Fall detection is an active research topic due to the need to prevent accidents from occurring among increasingly aged population in the world. Fall accident is not only harmful to elderly physical health but also will leave side effects, such as emotional trauma, to their daily life because of fear of falling again. Most fall-related research only focused on proposing methodologies to identify whether fall accidents occurred. This study approaches from analyzing elderly daily activities that may cause fall accidents. Interval type-2 fuzzy models are proposed to automatically detect elderly activity patterns. A multilayer detection system is devised to further identify elderly activity types. Signal Vector Magnitude (SVM) and Signal Magnitude Area (SMA) methods are used to discriminate fall activities from moderate and jog ones so that detection effort can be further simplified. Experimental results reveal that the proposed system can correctly identify fall activities. As for normal walking and jog activities the accuracy rates are higher than 80%.

June 2012, 11% of the overall population in Taiwan was over the age of 65. This ratio is higher than the average figure for the United Nations (8%). Critical issues concerning elderly in healthcare include fall detection, loneliness prevention and retard of obliviousness. In this study we design type-2 fuzzy models that utilize smart phone tri-axial accelerometer signals to detect fall incidents and identify abnormal gaits among elderly. Once a fall incident is detected an alarm is sent to notify the medical staff for taking any necessary treatment. When the proposed system is used as a pedometer, all the tri-axial accelerometer signals are used to identify the gaits during walking. Based on the proposed type-2 fuzzy models, the walking gaits can be identified as normal, left-tilted, and right-tilted. Experimental results from type-2 fuzzy models reveal that the accuracy rates in identifying normal walking and fall over are 92.3% and 100%, respectively, exceeding what are obtained using type-1 fuzzy models.

Clustering analysis provides significant contributions to healthcare or medical service. However, relying only on one set of clusters obtained from employing a clustering algorithm, such as fuzzy c-means algorithm (FCM), with an arbitrary initialization may be not robust and accurate in data clustering. The cluster ensemble, the concept of combining multiple clusters produced by a cluster algorithm with several different initializations, can improve the robustness problem. When the outliers were taken into the ensemble may lead the final cluster ensemble to inaccurate results. Thus, outliers should be removed before merging different clusters. In this paper, an adapted FCM algorithm is proposed to detect and remove the outliers. The cluster ensemble framework will employ this adapted FCM algorithm to generate multiple sets of clusters by giving different initialization parameters. Then, a pair wise approach is used to combine those outlier-free clusters. The experimental results verify that the final clusters obtained from the proposed cluster ensemble framework are more robust.

Manually operated medical equipment, including drug infusion pumps, are often subject to input errors. Human operators copy data from a prescription into the relevant form field on the equipment panels. This process is error prone and time consuming. A computer supported memory aid is proposed where the user remembers phrases instead of value sequences. The proposed strategy speeds up the task of setting up medical equipment while reducing the chances of human errors.

Most outlier detection algorithms are proposed to discover outlier patterns from static databases. Those algorithms are infeasible for instant identification of outlier patterns in data streams that continuously arriving and unbounded data serve as the data sources in many applications such as sensor data feeding. In this paper an association rules based method is proposed to find outlier patterns in data streams. The presented work segments transactions from data streams and then finds approximate frequent itemsets with single data scan instead of requiring multiple scans. Based on the derived association rules some transaction can be identified as outliers if their outlier degrees are higher than a predefined threshold. The proposed method not only just finds the outlier patterns but also identifies the most possible items that induce the abnormal transactions in the data streams. Efficiency comparisons with frequent itemsets-based work are also done to verify the effectiveness of the proposed framework.

Conventional water quality monitoring and bird species observation were recorded using pen and paper. Moreover, experiments can usually only be conducted in limited areas due to the high cost of the water monitoring equipment. Consequently, not enough data could be collected to find the a biotic interactions between water quality factors and bird habitat. To resolve this problem, the proposed smart phone based system can transmit the water quality measurements and observed quantity of each bird species to the back-end server via a 3G network. This server functions as a ubiquitous information center where all measured and observed data are stored. Fuzzy C-means is applied to cluster water quality factors and bird species. The clustering results of bird quantity are then used to calculate the Simpson's diversity indices. Based on the results from the water quality factors and biodiversity index pattern trees (PTs) are constructed to find the a biotic interactions between them. The PTs can be further transformed into IF-THEN rules to provide references for researchers to conduct environmental protection. Experimental results are given to verify the applicability of the proposed system in finding the a biotic interactions.

Instant identification of outlier patterns is very important in modern-day engineering problems such as credit card fraud detection and network intrusion detection. Most previous studies focused on finding outliers that are hidden in numerical datasets. Unfortunately, those outlier detection methods were not directly applicable to real life transaction databases. Although a limited literature presented methods to find outliers in the transaction datasets, they did not address what really caused the transactions to become abnormal. In this paper, an improved framework is proposed to identify the outlier transactions as well as to find the most possible items that induce the abnormal transactions. Several definitions are defined as prerequisite for outlier detection. Efficiency comparisons with previous work are also done to verify the effectiveness of the proposed framework.

This paper describes a model that discovers association rules from a medical database to help doctors treat and diagnose a group of patients who show similar prehistoric medical symptoms. The proposed data mining procedure consists of two modules. The first is a clustering module that is based on a neural network, Adaptive Resonance Theory 2 (ART2), which performs affinity grouping tasks on a large amount of medical records. The other module employs fuzzy set theory to extract fuzzy association rules for each homogeneous cluster of data records. In addition, an example is given to illustrate this model. Simulation results show that the proposed algorithm can be used to obtain the desired results with a reduced processing time.

With the constant stress from work load and daily life people may show symptoms of melancholia. However, most people are reluctant to describe it or may not know that they already have it. In this paper a novel system is proposed to discover clues from patient's interaction with psychologist or from self-recorded voice or text messages. A user friendly interface is provided for patients to input text messages or record a voice file by mobile phones or other input devices. A speech-to-text conversion software is used to convert voice mails to simple text files in advance. Based on the text files, a data mining model is used to discover frequent keywords mentioned in the text or speech files. The association rules can be used to help psychologists diagnose patients' degree of melancholia. Experimental results show that the proposed system can effectively discover melancholia keywords.

In order to navigate autonomously, a mobile robot needs to build an environment map where the robot is navigating. Currently, the sensors are mounted on the robot to detect if the obstacles exist and then the map immediate surrounding of the robot is built to help for navigation path planning. The map created by this method is a local map that may cause global navigation problem which a global coverage map is needed to solve such a problem. In this study, a sensor network is deployed for building global environment map. All the sensor locations are assumed known. The navigation space is divided into grids and a grid is to be detected if obstacles exist by one or a number of sensors. Fuzzy set concept is used to introduce a tool useful for sensor perception. Those sensors work as a team to explore all the space and then the global fuzzy map is constructed. The experiments show that the fuzzy map is more practical and helps the path planning problem to be solved more efficiently.

Information retrieval systems should provide users quick access to desired information. There are no established ways for inexperienced users to explicitly express queries for retrieving images from ecological databases. This study proposes an entropy-based feature selection strategy for finding images of interest from databases. Six visual features are used to represent birds, and hence used to formulate search queries. The proposed method is tested on a real world bird database and the experimental results demonstrate the effectiveness of the presented work.

Several research institutions and governmental departments provide ocean images for research purposes. For example, Argo, a worldwide ocean research organization, produces ocean salinity and temperature images and researchers can download those images from the Internet. One may build an image system to store ocean images and retrieve them later for further research, for example, to predict future salinity or temperature variation. Image retrieval technology is therefore important. This paper describes an ocean image retrieval system based on content-based image retrieval. Currently, content-based image retrieval technology does not exploit high-level semantics, and it is hard to obtain predictive information from retrieved images. Our improvement involves a spatial reference method that is used to help get the spatial relationships between objects for a certain image. This allows the spatial semantics between the query image and images in database to be considered. Spatial association rules are also mined and are subsequently used as a basis for retrieving additional images. As the spatial semantics in both the query image and spatial association rules, the retrieved images are more accurate. The experimental results verify that the system effectively predicts the occurrence of salinity or temperature variations.

A novel and more effective region-based image retrieval strategy is presented based on semantic ontology. An unsupervised segmentation algorithm splits images into regions that are subsequently used as basis by the ontology-based strategy. The approach comprises three stages, namely automatic region generation, categorization and ontology construction. When receiving a query for a specific object, the search engine will, in addition to conventionally matched images, also find candidates through the semantic ontology using low level features. The proposed approach can thus find a richer set of related candidate images than traditional image retrieval approaches. This strategy is particularly useful for vague queries encountered by inexperienced users that are not trained in searching for images by the means of low-level features. The experimental results demonstrate the effectiveness of the proposed approach.

In this paper, an efficient two-stage approach is proposed for content-based image retrieval (CBIR). In establishing the database, the features of an image are extracted from its color histograms and discrete cosine transform (DCT) coefficients. To improve the retrieval performance, the quantization technique is applied to quantize the vector of color histograms such that the feature space is partitioned into a finite number of grids, each of which corresponds to a grid code (GC). At the first stage, a reduced set of candidate images which have the same GC (or adjacent GCs) as that of the query image is obtained. At the second stage, the remaining candidates are examined by using grey relational analysis on the significant DCT coefficients. The experimental results show that the proposed approach leads to a fast retrieval with good accuracy.

Ocean salinity and temperature variations have been observed for decades to clarify their effect to global climate changes. Data mining techniques are effective in extracting implicit and useful information from large databases. Discovering salinity and temperature variation patterns from Argo ocean data will in turn help reveal the spatio-temporal relationship between salinity and temperature variations. However, some of the discovered patterns are trivial because they are already known to the oceanographer. In this study, the water mass (a water body with the same salinity and temperature), the mined salinity and temperature patterns and an entropy importance measure are combined to discriminate important patterns from trivial patterns. This study measures both the patterns with variations in both antecedent and consequent parts that belong to separate clusters, and that belong to the same cluster. A pattern is classified as important if its importance measure exceeds a predefined threshold. The important patterns are transformed into fuzzy rules in a fuzzy inference model to obtain more accurate salinity and temperature variation predictions. Simulation results verify the effectiveness of the proposed model.

License plate recognition systems have been used extensively for many applications including parking lot management, tollgate monitoring, and for the investigation of stolen vehicles. Most researches focus on static systems, which require a clear and level image to be taken of the license plate. However, the acquisition of images that can be successfully analyzed relies on both the location and movement of the target vehicle and the clarity of the environment. Moreover, only few studies have addressed the problems associated with instant car image processing. In view of these problems, a real-time license plate recognition system is proposed that recognizes the video frames taken from existing surveillance cameras. The proposed system finds the location of the license plate using projection analysis, and the characters are identified using a back propagation neural network. The strategy achieves a recognition rate of 85.8% and almost 100% after the neural network has been retrained using the erroneously recognized characters, respectively.

An improved shape based image retrieval strategy based on gray relational analysis and association rules is proposed. The choice of a suitable object representation and retrieval scheme is essential for efficient retrieval. In addition, a two-stage relevance feedback mechanism based on the GM(1, N) method and association rules is incorporated to improve the retrieval accuracy. The GM(1, N) method is used to build the re-query example for subsequent retrievals. The retrieval log files stored on the server are used for offline mining of association rules. The association rules mined from users' retrieval history can further reveal users' image searching behavior. The effectiveness of the proposed model is demonstrated on the FISH dataset.

In this paper, a two-stage content-based image retrieval (CBIR) approach is proposed to improve the retrieval performance. To develop a general retrieval scheme which is less dependent on domain-specific knowledge, the discrete cosine transform (DCT) is employed as a feature extraction method. In establishing the database, the DC coefficients of Y, U and V components are quantized such that the feature space is partitioned into a finite number of grids, each of which is mapped to a grid code (GC). When querying an image, at coarse classification stage, the grid-based classification (GBC) and the distance threshold pruning (DTP) serve as a filter to remove those candidates with widely distinct features. At the fine classification stage, only the remaining candidates need to be computed for the detailed similarity comparison. The experimental results show that both high efficacy and high efficiency can be achieved simultaneously using the proposed two-stage approach.

Providing users with easy access to interesting information is one important goal of information retrieval. However, traditional information retrieval techniques are not suitable for ecological database queries since users are unable to clearly describe the target features. To overcome the problem, an information retrieval system running on handheld devices allowing users to issue queries according to their perceptions or impressions of the target bird is proposed. Six visual features allow users to express their visual perception of a bird and the query will be formulated based on the chosen features. Texture descriptions are also used when searching for the target in the bird database. A bird-information ontology based on the biodiversity is exploited to extend the search space. The methodology is illustrated in this paper. The experimental results demonstrate the effectiveness of the strategy and its applicability in other domains.

The ARGO project archives huge quantities of upper ocean salinity/temperature time series measurements that are related to climate issues such as global warming. Fuzzy inter-transaction association rules are derived from ARGO data using a reduced prefix-projected itemset algorithm that has a small space and time complexity. After mining the frequent 1-itemsets the proposed algorithm exploits a reduced prefix projection strategy to extract the frequent inter-itemsets. Based on the extracted fuzzy inter-transaction association rules a fuzzy inference model is proposed for identifying salinity/temperature anomalies. Experimental results verify that the proposed model is effective in predicting the occurrence of abnormal salinity/temperature variations.

A strategy for locating television (TV) commercials in TV programs is proposed. Based on the observation that most TV commercials do not have subtitles, the first stage exploits six subtitle constraints and an adaptive neurofuzzy inference system model to determine whether a frame contains a subtitle or not. The second stage involves locating the mark-in/mark-out points using a genetic algorithm. An interactive user interface allows users to efficiently identify and fine-tune the exact boundaries separating the commercials from the program content. Furthermore, erroneous boundaries are manually corrected. Experimental results show that the precision rate and recall rates exceed 90%.

In this paper a load-balancing strategy for distributed Web-applications is proposed. The approach assumes uneven traffic with extreme peaks that cannot be served with a given infrastructure. The approach presented herein is based on distributing the load temporally in time. In online Web-based applications the perceived response time is more important than the actual response time and various strategies for manipulating the perception of response time are discussed. The technique is relatively easy to incorporate with existing Web-applications using modern Web-technology such as AJAX. Simulation results demonstrate the effectiveness of the approach

This paper presents a content-based image retrieval method based on the discrete wavelet transform (DWT). Due to the superiority in multiresolution analysis and spatial-frequency localization, the DWT is used to extract wavelet features (i.e., approximations, horizontal details, vertical details, and diagonal details) at each resolution level. Based on the observation that the YUV color space is rather effective in terms of the extraction of color features, each image is first transformed from the standard RGB color space to the YUV space, and then each component (i.e., Y, U, and V) of the image is further transformed to the Wavelet domain. In the image database establishing phase, the wavelet coefficients of each image are stored; in the image retrieving phase, the system compares the most significant wavelet coefficients of the Y, U, and V components of the query image with those of the images in the database, coupled with the weight factors assigned by users, and find out the matches based on the users' interested features. Experimental results demonstrate the effectiveness of our system.

A novel data mining techniques for finding interesting spatial-temporal patterns in ocean data is presented. The data consist of time series measurements of upper ocean science variables (e.g., salinity and temperature). Extracting interesting patterns from ocean variables is of important for understanding the relationship between ocean salinity/ temperature structures and climate variability. Association rules mining is applied in the search for these spatial-temporal patterns. Most traditional data mining models focus on mining association rules among attributes within one transaction. For example, if salinity rose, then temperature rises. However, people may be interested in discovering additional relations among transactions and take context such as time or location into consideration. An example of such a rule might be "if the salinity in area A rose from 5% to 7%, then the temperature in area B will rise from 0% to 2.5% in the next month." In this case, the associated salinity/temperature variations among different locations and days are revealed. To overcome these issues, a multi-dimensional inter-transaction association rules mining framework was developed. Unlike other mining algorithms that suffer from a large number of inter-transaction items, the proposed Apriori-like method treats each event from the ocean science data as a transaction and applies the MBC (minimum bounding cube) to form inter-transactions within the maxspan. Since there is no need to slide the maxspan window, the proposed method is easy to implement and it is computationally efficient.

A general input paradigm for blind and visually impaired users is proposed. Input is classified into three categories, namely input of text using Braille chords, input of digits and navigation and editing of text. A numpad input device is used for all the three modes of input. The required mode is automatically detected by analyzing the spatial and temporal properties of the device usage. Braille input chords are resolved dynamically and are therefore not bound to specific keys. Low-cost and widely available standalone numeric keypads are used as input devices.

Automatically retrieving images through their low-level visual features has become one of the challenging areas of research recently. Among those distinguishing features, the texture features are one of the main themes in content-based image retrieval (CBIR). In this paper, we propose a novel technique to extract dominant features of images in block-DCT domain. The image is first converted to YUV color space and divided into four subblocks. The Y-component in each subblock is then transformed into DCT coefficients, some regions of which characterize different directional texture feature of that subblock. The directional textures in all subblocks are concatenated together as a single feature vector and used for indexing and retrieval of images. The experimental results show that using proper size of block-DCT to emphasize the regional properties of an image while maintaining its global view performs well in CBIR.

Most of the previous studies on mining association rules focused on mining Boolean intra-transaction associations, i.e., the association rules among binary attributes within the same transaction where the notion of the transaction could be the items bought by the same customer. In this paper, we deal with the problem of mining association rules in databases containing quantitative attributes to discover the associations among different transactions. An example of such an association might be "if company A's stock closing price goes up 1% to 3%, company B's; price goes up 2% to 4% the next day." In this case, no matter whether we treat company or day as the unit of transaction, the associated items belong to different transactions. However, a problem is caused by sharp boundary in this example. For instance, if A's stock closing price goes up only 0.99%, the example we illustrate is not applicable to predict company B's; stock price the next day. The fuzzy set concept can help us tackle this kind of problem since fuzzy sets provide a smooth transition between member and non-member of a set. Besides, to mine inter-transaction association rules from the 1-dimensional database, a sliding window concept is introduced. Each sliding window in the database forms a mega-transaction and the associations from these mega-transactions can thus be found. Our algorithm first employs fuzzy set to map quantitative attributes into fuzzy attributes and an a priori-like method is developed to find inter-transaction fuzzy association rules. As compared with conventional methods, more useful results can be found from the proposed fuzzy association rules

A novel strategy for employing schedules obtained using standard static scheduling algorithms in a battery powered multiprocessor environment is investigated. The strategy is able to dynamically respond to deteriorating batteries and operate with fewer processors according to the battery levels of the system. The nature of the proposed approach allows poorly performing batteries to recover through self charge. The strategy therefore maximizes the battery life and the operation time of the device.

In this paper, we present a two-stage classification approach to recognize the characters in the rare books transcribed by ancient calligraphers. The first stage is coarse classification which uses grid code transformation (GCT) method to quantize the most significant discrete cosine transform coefficients into a finite number of grids. On classifying an unknown character, a reduced set of candidate classes can be retrieved from the corresponding grid code. The second stage is fine classification, which uses a statistical mask-matching method to identify the individual target in the set given by the first stage. In the training phase, we generate one positive mask and one negative mask for each distinct class of characters. Therefore, an unknown character can be recognized by finding the prototype character whose masks are best fitted to it. Experiments were conducted for recognizing handwritten characters in Chinese paleography and showed that our approach performs well in this application domain.

There are two problems as we use conventional Boolean association rules mining algorithm to discover temporal association rules over the stock market to predict stock price variation. The first problem is that the discovered rules only consider associations between the presence and absence of variations of stock prices and the second problem is that the associations among stock price variations are within the same transaction day. For example, if stock A raises, then stock B raises the same day. This Boolean temporal association rule reveals no information of quantitative variations of stock prices and can only predict price trend in the same day. In this paper, we deal with the problem of mining temporal association rules in stock databases containing quantitative price variations to discover the associations among different transactions day. Our algorithm first employs data discretization concept to partition quantitative attributes into intervals and an adaptive a priori method that cooperates with time sliding window concept and prefix tree is developed to find quantitative temporal association rules. An example of such a rule might be "if stock A price variation raised 5% to 7% and stock B raised 2.5% to 5% the same day, then stock C will raise 0% to 2.5% in the next two days." In this case, the stock price variation is taking into consideration and the associated stock price variations belong to different transaction days. As compared with conventional methods, more useful results can be found from the proposed quantitative temporal association rules.

We present a novel method to segment objects in images based on the similarity measurement of fuzzy gray level technique in this paper. In our model, we classify the processing steps into three stages. First, we utilize the attributes of luminance and chromaticity components of HLS color coordinate system to form a fuzzy gray level. These attributes can describe the relationship between different frequent colors and the image can be transferred to smooth gray level, which can capture the objects in images. Second, we reduce the gray levels of image pixels to lower gray levels to speed up computation. Third, we label each root pixel based on a similarity measurement. We perform a sliding window to move from one block to the next one. The similarity of the two root pixels blocked by the sliding window depends on their neighboring pixels. Via the similarity computation, we assign a label number to the root pixels. We generate objects from grouping different labels. The image data are classified by fuzzy gray level technique and the objects are segmented from images. According to the simulation results, our model shows the efficiency and effectiveness for image segmentation.

Traditional information retrieval techniques are quantitative approaches. That is, if the only concern is to find information that completely or partially matches users' queries. However, the retrieval task is unsatisfactory if the definite forms are not easily or possibly represent the semantic contents of the queries. Thus, we propose a fuzzy information retrieval model that can "understand" users' queries especially when the users cannot clearly describe the part or the whole features of their query specifications. User query which is viewed as a semantic entry, could belong to the multiple semantic categories and by introducing two fuzzy measure degrees, centrality and intensity, our model is capable of dealing with the ambiguity in user query. The matching policy is based on the combining centrality distance and the intensity distance between the query and the targets in database. The total distance is taking into account the confidence values of all the considered features. Since the model is qualitative approach, the system can capture what the users' can hardly express. The system can "see" what users' interest are, even if the user cannot or don't know how to explicitly express what they have remembered in the form of queries.

Based on the available transaction records, we use AprioriTid model to derive the association rules from large database. We then exploit association rules to establish an initial fuzzy inference model. A novel tuning method is proposed to adjust the fuzzy model such that every association rule from data mining model can in turn help us recommend the most appropriate products to the prospective customers. By combining the Larsen's inference method and gradient descent method, we derive a systematic approach to refine the fuzzy model. Thus, a new adjusting method, i.e., Larsen-like, is proposed in this paper. How to derive the association rules from large database, how to apply the derived rules to establishing a fuzzy inference model, and how to optimize the fuzzy model are illustrated by simple examples.

A fuzzy inference model is generated to fulfill the personalized service through mining the association rules from a large database in this paper. Instead of just considering whether interesting items have appeared in the same transaction, we also investigate other aspects, such as the purchased quantity, associated with the items. Based on the proposed model, our system can predict which items should be recommended to the prospective customers to realize the personalized service. How to derive the association rules from large database and how to apply the derived rules to establishing a fuzzy inference model are illustrated by simple examples.

A novel fast tuning algorithm is proposed to expedite the converging process in the parameter identification of fuzzy models. In order to improve the disadvantages of the time-consuming gradient descent method, the principle of this new algorithm is only to tune the consequent parts of the fuzzy rules. The membership functions of the fuzzy model remain unchanged. The proposed tuning method is applicable to two different types of fuzzy rules. Some simulation results are given to verify that the proposed method can converge speedily and have better inference capability than conventional methods.

Assume X/sup #/ is an extension set in the domain U. Let X/sup A/, X/sup B/ be the positive domain of X/sup #/ in which X/sup A//spl sub/X/sup #/, X/sup B//spl sub/X/sup #/, and X/sup A//spl ne/X/sup B/. X/sup A/={x/sub 1//sup A/, x/sub 2//sup A/, ..., x/sub n//sup A/}, X/sup B/={x/sub 1//sup B/, x/sub 2//sup B/, ..., x/sub n//sup B/}. /spl forall/x/sub 1//sup A/, x/sub 1//sup B//spl isin/N/sup +/, |X/sup A/|/spl ges/4, |X/sup B/|/spl ges/4. By using the accumulated generating operation, we can obtain a mathematical model from X/sup A/ and X/sup B/. From the mathematical model we also derive the sets of X'/sub A/ and X'/sub B/. Both X'/sub A/ and X'/sub B/ are the extended domains of X/sup #/. There exist polynomials p/sup A/(x), p/sup B/(x); and p/sup AB/(x) from X'/sub A/, X'/sub B/, and X/sup AB/, respectively. Based on the p/sup A/(x), p/sup B/(x), and p/sup AB/(x), we propose a new public cryptosystem. In the presented system, we have a public cryptography system from generated data in extension set X/sup #/, a public cryptosystem from two generated data in extension set X/sup #/, and a public cryptosystem from hybrid generated data in extension set X/sup #/. The proposed model is a secure system. Both sides in the communication system can depend on the proposed cryptosystem to fulfill the cryptography of the to-be-transmitted file and de-cryptograph of the cryptograph.

Adjusting the membership functions to satisfy one pattern may deteriorate the inference outcomes of the others. This incompatible issue can be retarded by the extension theory. A novel extension-based fuzzy modeling method, which differs from the traditional fuzzy inference, is proposed. Instead of directly applying the given data to building the fuzzy model, the given data are transformed to another domain by a sigmoidal function to obtain a better fuzzy model. We also define the extended correlation functions to relate the data with the fuzzy sets. During the refining process, the extended fuzzy model, which considers the positive and negative sets simultaneously, is adjusted by the gradient descent method. Simulation results from both single-input-single-output and double-input-single-output systems verified that better results than the conventional methods can be obtained.

With the advent of the Internet and WWW in the late 1990s, intelligent systems have found another application area. In the Internet-based system, how to search for the desired information in a short time is very important for the agent, search engine and data mining systems. Thus, we try to design an intelligent system with the characteristics of low complexity, quick convergence and low output error for both academia and industry. We use grey relational analysis to select more important input variables to establish a simplified fuzzy model. Then, we exploit the concepts of extension theory to adjust the fuzzy model during the parameter identification to expedite the tuning process. Finally, the proposed extension-based fuzzy model is applied to implementing an intelligent information retrieval system for the search engine to stress the model's applicability.

A new extended fuzzy model is proposed in this paper. The newly established extension theory is integrated into the conventional fuzzy system to enhance the reasoning capability. Not only the sample data located at the classical fuzzy set have effect to adjust the mapped membership function, but also the data in the neighboring regions are simultaneously considered. Simulation results from the extended fuzzy model are given and the comparisons with the results from ellipsoidal fuzzy method and other approaches are also made.

In this paper several grey-based models are applied to temperature prediction problems. Standard normal distribution, linear regression, and fuzzy techniques are respectively integrated into the grey model to enhance the embedded GM(1, 1), a single variable first order grey model, prediction capability. The original data are preprocessed by the statistical method of standard normal distribution such that they will become normally distributed with a mean of zero and a standard deviation of one. The normalized data are then used to construct the grey model. Due to the inherent error between the predicted and actual outputs, the grey model is further supplemented by either the linear regression or fuzzy method or both to improve the prediction accuracy. Results from predicting the monthly temperatures for two different cities demonstrate that each proposed hybrid methodology can somewhat reduce the prediction errors. When both the statistics and fuzzy methods are incorporated with the grey model, the prediction capability of the hybrid model is quite satisfactory. We repeat the prediction problems in neural networks and the results are also presented for comparison.

A general form of input-destination distribution matrix increases state space exorbitantly, thus making any buffer at every state statistically different from another. Certain specific forms of input-destination distribution matrix to which many real-life cases may conform, are analyzed. The idea called decomposition is applied here for specific nonhomogeneous flows. State space is not allowed to increase significantly; also the reduction in network size at successive stages is utilized to increase the computational efficiency.>