The robustness, security, and safety of artificial intelligence (AI) systems have become growing concerns, particularly as deep learning models are increasingly deployed in critical applications. Among emerging threats, backdoor attacks pose a serious risk by embedding hidden malicious behaviors into otherwise well-performing models. Although recent advances in detection techniques have improved defenses for computer vision systems, our findings demonstrate that even simple but carefully designed poisoning strategies can successfully evade these defenses. In this paper, we introduce GhostBackdoor, a novel backdoored model trained using a custom loss function and targeted data augmentation. The proposed loss function aligns neuron activations between clean and poisoned inputs, effectively masking activation anomalies, while the augmentation enforces strict location- and pattern-specific triggers that activate the backdoor only under specific conditions. After training, the model maintains behavior indistinguishable from a clean model unless exposed to the designated trigger with the specific pattern and at the designed locations. We evaluate GhostBackdoor against a broad range of leading defense mechanisms, most of which fail to detect the implanted backdoor. Our results highlight how the vast hypothesis space of deep learning models can be exploited to conceal malicious activations, underscoring the need for more robust security strategies in AI-driven systems, including those used in Internet of Things (IoT) applications.

This paper proposes SEER, a novel backdoor detection algorithm for vision-language models, addressing the gap in the literature on multi-modal backdoor detection. While backdoor detection in single-modal models has been well studied, the investigation of such defenses in multi-modal models remains limited. Existing backdoor defense mechanisms cannot be directly applied to multi-modal settings due to their increased complexity and search space explosion. In this paper, we propose to detect backdoors in vision-language models by jointly searching image triggers and malicious target texts in feature space shared by vision and language modalities. Our extensive experiments demonstrate that SEER can achieve over 92% detection rate on backdoor detection in vision-language models in various settings without accessing training data or knowledge of downstream tasks.

Federated Learning (FL) has been adopted in practical network applications and plays a critical role. As FL allows participants to contribute to the global model by training locally with private data, it is known particularly vulnerable to neural backdoor attacks. This paper proposes a new defense, ScanFed, against neural backdoor attacks to FL systems. It leverages the synchronous nature of FL to effectively single out malicious neuron candidates and further validate if they indeed hijack the model's behaviors. Compared to existing neural backdoor defenses, ScanFed has the following distinct properties. First, it is extremely computation-friendly that is six orders of magnitude faster than state-of-the-art behavior-based backdoor defenses, rendering it highly suitable for large-scale FL systems. Second, it inherits the precise nature of behavior-based backdoor detection, making it significantly more effective than similarity-based defenses against advanced attacks. Third, it is robust to biased models uploaded by clients with non-IID (Independent and Identically Distributed) data, which is very common in practical FL systems. In addition, it is a plug-n-play scheme that can be seamlessly integrated into existing FL systems. To the best of our knowledge, this is the first behavior-based defense that enables scalable, efficient and accurate neural backdoor detection of FL systems in non-IID scenarios. This work delivers a ScanFed prototype and fully tests it in various settings of datasets, neural architectures, and backdoor attacks. The experiments demonstrate ScanFed achieves competitive accuracy and minimal detection time.

Choroidal neovascularization (CNV) is a typical symptom of age-related macular degeneration (AMD) and is one of the leading causes for blindness. Accurate segmentation of CNV and detection of retinal layers are critical for eye disease diagnosis and monitoring. In this paper, we propose a novel graph attention U-Net (GA-UNet) for retinal layer surface detection and CNV segmentation in optical coherence tomography (OCT) images. Due to retinal layer deformation caused by CNV, it is challenging for existing models to segment CNV and detect retinal layer surfaces with the correct topological order. We propose two novel modules to address the challenge. The first module is a graph attention encoder (GAE) in a U-Net model that automatically integrates topological and pathological knowledge of retinal layers into the U-Net structure to achieve effective feature embedding. The second module is a graph decorrelation module (GDM) that takes reconstructed features by the decoder of the U-Net as inputs, it then decorrelates and removes information unrelated to retinal layer for improved retinal layer surface detection. In addition, we propose a new loss function to maintain the correct topological order of retinal layers and the continuity of their boundaries. The proposed model learns graph attention maps automatically during training and performs retinal layer surface detection and CNV segmentation simultaneously with the attention maps during inference. We evaluated the proposed model on our private AMD dataset and another public dataset. Experiment results show that the proposed model outperformed the competing methods for retinal layer surface detection and CNV segmentation and achieved new state of the arts on the datasets.

Deep learning techniques have proven effective in many applications, but these implementations mostly apply to data in one or two dimensions. Handling 3D data is more challenging due to its irregularity and complexity, and there is a growing interest in adapting deep learning techniques to the 3D domain. A recent successful approach called MeshCNN consists of a set of convolutional and pooling operators applied to the edges of triangular meshes. While this approach produced superb results in classification and segmentation of 3D shapes, it can only be applied to edges of a mesh, which can constitute a disadvantage for applications where the focuses are other primitives of the mesh. In this study, we propose face-based and vertex-based operators for mesh convolutional networks. We design two novel architectures based on the MeshCNN network that can operate on faces and vertices of a mesh, respectively. We demonstrate that the proposed face-based architecture outperforms the original MeshCNN implementation in mesh classification and mesh segmentation, setting the new state of the art on benchmark datasets. In addition, we extend the vertex-based operator to fit in the Point2Mesh model for mesh reconstruction from clean, noisy, and incomplete point clouds. While no statistically significant performance improvements are observed, the model training and inference time are reduced by the proposed approach by 91% and 20%, respectively, as compared with the original Point2Mesh model.

We report a new neural backdoor attack, named Hibernated Backdoor, which is stealthy, aggressive and devastating. The backdoor is planted in a hibernated mode to avoid being detected. Once deployed and fine-tuned on end-devices, the hibernated backdoor turns into the active state that can be exploited by the attacker. To the best of our knowledge, this is the first hibernated neural backdoor attack. It is achieved by maximizing the mutual information (MI) between the gradients of regular and malicious data on the model. We introduce a practical algorithm to achieve MI maximization to effectively plant the hibernated backdoor. To evade adaptive defenses, we further develop a targeted hibernated backdoor, which can only be activated by specific data samples and thus achieves a higher degree of stealthiness. We show the hibernated backdoor is robust and cannot be removed by existing backdoor removal schemes. It has been fully tested on four datasets with two neural network architectures, compared to five existing backdoor attacks, and evaluated using seven backdoor detection schemes. The experiments demonstrate the effectiveness of the hibernated backdoor attack under various settings.

Remote sensing integrated with deep learning (DL) improves wildfire assessment. The research has been done to scrutinize the areas affected by disastrous wildfires in Yunnan using DL. Wildfire identification and demarcation of the affected area have been limited to primitive thresholding and outdated machine learning classification techniques. Therefore, the research work incorporated DL in the wildfire scrutiny, and several of the most important considerations are investigated. The proposed research objective is to exploit the recent advancement of capsule-based DL together with the wildfire domain. The proposed dense structure provides highly efficient detection and segmentation of the burned area (BA). The BA dense capsule network (BA_EnCaps) is employed to extract and localize the burned zone with an overall accuracy of 98%. The model is evaluated quantitatively using accuracy, binary_cross-entropy, dice_loss, and mean square error (mse). The research aims to utilize the segmentation model to estimate the BA with great results. BA-EnCaps shows excellent accuracy in discriminating the spectral indices for the burned zone. The proposed method surpasses other segmentation benchmark techniques (U-Net, U-Net3p, SegCaps, Deep U-Net, and U-Net+) by substantially lessening the computing power. Finally, BA_EnCaps is compared with standard segmentation techniques and shows that DL-based models can assess wildfire better than conventional algorithms.

The statistical data from the National Council on Aging indicates that a senior adult dies in the US from a fall every 19 minutes. The care of elderly people can be improved by enabling the detection of falling events, especially if it triggers the pneumatic actuation of a protective airbag. This work focuses on detecting impending fall risk of senior subjects within the geriatric population, towards a planned approach to mitigating fall injuries through pneumatic airbag deployment. With the widespread adoption of wearable sensors, there is an increased emphasis on fall prediction models that effectively cope with accelerometry signal data. Fall detection and gait classification are challenging tasks, especially in differentiating falls from near falls. We propose to apply attention to the deep neural network (DNN) analysis of acceleration data where a fall is known to have occurred. We take the maximum value of the sensor signals to define the observation window of the detector. Powered by a transformer DNN with word embedding, attention networks have achieved a state-of-the-art in natural language processing (NLP) tasks. Besides the success of the transformer for efficiently processing long sequences, it supports parallel computing with fast computation. In this paper, we propose a novel transformer attention network for gait analysis of fall detection modeling with Time2Vec positional encoding- founded on a Masked Transformer Network. Using our dataset, we demonstrate that the proposed approach achieves better specificity and sensitivity than the present models.

This paper reports a new clean-label data poisoning backdoor attack, named Invisible Poison, which stealthily and aggressively plants a backdoor in neural networks. It converts a regular trigger to a noised trigger that can be easily concealed inside images for training NN, with the objective to plant a backdoor that can be later activated by the trigger. Compared with existing data poisoning backdoor attacks, this newfound attack has the following distinct properties. First, it is a blackbox attack, requiring zero-knowledge of the target model. Second, this attack utilizes "invisible poison" to achieve stealthiness where the trigger is disguised as `noise', and thus can easily evade human inspection. On the other hand, this noised trigger remains effective in the feature space to poison training data. Third, the attack is practical and aggressive. A backdoor can be effectively planted with a small amount of poisoned data and is robust to most data augmentation methods during training. The attack is fully tested on multiple benchmark datasets including MNIST, Cifar10, and ImageNet10, as well as application specific data sets such as Yahoo Adblocker and GTSRB. Two countermeasures, namely Supervised and Unsupervised Poison Sample Detection, are introduced to defend the attack.

Rukban is a desert area crossing the border between Syria and Jordan, and thousands of Syrian refugees fled into this area since the Syrian civil war in 2014. In the past few years, the number of refugee shelters for the forcibly displaced Syrian refugees in this area has increased rapidly. Estimating the location and number of refugee tents has become a key factor to maintain the sustainability of the refugee shelter camps. Manually counting the shelters is labor-intensive and sometimes prohibitive given the large quantities. In addition, these shelters/tents are usually small in size, irregular in shape, and sparsely distributed in a very large area and could be easily missed by the traditional image-analysis techniques, making the image-based approaches also challenging. In this letter, we proposed a deep fully convolutional neural network (FCN) model to extract automatically the refugee shelters/tents in the worldview-2 (WV-2) satellite images. In addition, we transferred knowledge in the pretrained VGG-16 model to improve the detection accuracy and network training convergence. We compared the proposed approach with the traditional spectral angle mapper (SAM) method, deep convolutional neural network (CNN) models, and the mask Region-based CNN (R-CNN) model. The experimental results show that the FCN model improved the overall accuracy by 4.49%, 3.54%, and 0.88% compared with the CNNs, SAM, and mask R-CNN models, and improved the precision by 34.61%, 41.99%, and 11.87%, respectively.

Seagrass form the basis for critically important marine ecosystems. Previously, we implemented a deep convolutional neural network (CNN) model to detect seagrass in multispectral satellite images of three coastal habitats in northern Florida. However, a deep CNN model trained at one location usually does not generalize to other locations due to data distribution shifts. In this paper, we developed a semi-supervised domain adaptation method to generalize a trained deep CNN model to other locations for seagrass detection. First, we utilized a generative adversarial network loss to align marginal data distribution between source domain and target domain using unlabeled data from both data domains. Second, we used a few labelled samples from the target domain to align class specific data distributions between the two domains, based on the contrastive semantic alignment loss. We achieved the best results in 28 out of 36 scenarios as compared to other state-of-the-art domain adaptation methods.

Seagrass form the basis for critically important marine ecosystems. Previously, we implemented a deep convolutional neural network (CNN) model to detect seagrass in multispectral satellite images of three coastal habitats in northern Florida. However, a deep CNN model trained at one location usually does not generalize to other locations due to data distribution shifts. In this paper, we developed a semi-supervised domain adaptation method to generalize a trained deep CNN model to other locations for seagrass detection. First, we utilized a generative adversarial network (GAN) loss to align marginal data distribution between source domain and target domain using unlabeled data from both data domains. Second, we used a few labelled samples from the target domain to align class specific data distributions between the two domains, based on the contrastive semantic alignment loss. We achieved the best results in 28 out of 36 scenarios as compared to other state-of-the-art domain adaptation methods.

The land resource is becoming scarcer and scarcer for a rapidly developing city. Thus, the land price assessment is important for the government to auction the land appropriately. In the paper, we introduced the deep neural network to evaluate the land price, taking the Shenzhen city in China as a case. Firstly, twenty influencing factors and land price data were gathered. Then, Shenzhen city was segmented into many grids with a size of 300 × 300 m. Secondly, the land price of each grid was derived with Kriging approach based upon the samples of land price. And the twenty influencing factors was quantified. Thirdly, the land price data and influencing factors were partitioned into training and testing datasets with the ratio of 8:1, and the training data were utilized to train the deep neural network based on regression analysis and classification with different hidden layers. Finally, the results were analyzed, and the deep neural network with the highest accuracy was selected as the optimum model. Therefore, our proposed method is an efficient approach to evaluate the land price with deep neural network.

Hyperspectral image is usually composed of hundreds of bands rich of spatial and spectral information. And this is an advantage for the common remotely sensed data. Thus, the classification of hyperspectral image could be of great value. However, the dimensionality of hyperspectral image may lead to the curse of dimensionality phenomenon when it is directly used for land use classification or other applications, making it difficult to be utilized effectively. In this paper, we presented a novel classification framework with capsule network based on the spectral and spatial information of hyperspectral images. At first, we use principal components analysis (PCA) to reduce the dimensionalities of hyperspectral image. Then, we use the capsule network to classify hyperspectral image. Our experimental result showed the novel classification framework is more efficient than other six popular methods. Therefore, the capsule network method is robust for hyperspectral image classification.

In recent years, the big data industry chain has become more mature. Analyzing and managing cities by utilizing various big data in cities has become a hot research topic. Urban functional regions discovering is one of the important applications. The mainstream in urban functional regions discovering are probabilistic topic models, such as latent Dirichlet allocation (LDA) based topic model, which seeing the regions as documents and their functions are their topics. These methods require feature engineering by hand, which will construct features of limited expressiveness. To overcome these methods' shortcomings, we introduced a deep learning topic model called document neural autoregressive distribution estimation (DocNADE) into urban functional regions mining. And we did an experiment to test its effect. The experimental result shows that this DocNADE framework has achieved a considerable result in urban function inference compared with Dirichlet Multinomial Regression (DMR) based topic model which is a state of the art of urban functional regions discovering.

This work proposes an innovative approach, named CapJack, to detect in-browser malicious cryptocurrency mining activities by using the latest CapsNet technology. To the best of our knowledge, this is the first work to introduce CapsNet to the field of malware detection through system behavioral analysis. It is particularly effective to detect malicious miners under multitasking environments where multiple applications run simultaneously. Experimental data show appealing performance of CapJack, with a detection rate of as high as 87% instantly and 99% within a window of 11 seconds.

Public digital media can often mix factual information with fake scientific news, which is typically difficult to pinpoint, especially for non-professionals. These scientific news articles create illusions and misconceptions, thus ultimately influence the public opinion, with serious consequences at a broader social scale. Yet, existing solutions aiming at automatically verifying the credibility of news articles are still unsatisfactory. We propose to verify scientific news by retrieving and analyzing its most relevant source papers from an academic digital library (DL), e.g., arXiv. Instead of querying keywords or regular named entities extracted from news articles, we query domain knowledge entities (DKEs) extracted from the text. By querying each DKE, we retrieve a list of candidate scholarly papers. We then design a function to rank them and select the most relevant scholarly paper. After exploring various representations, experiments indicate that the term frequency-inverse document frequency (TF-IDF) representation with cosine similarity outperforms baseline models based on word embedding. This result demonstrates the efficacy of using DKEs to retrieve scientific papers which are relevant to a specific news article. It also indicates that word embedding may not be the best document representation for domain specific document retrieval tasks. Our method is fully automated and can be effectively applied to facilitating fake and misinformed news detection across many scientific domains.

Human voice could be significantly affected by neoplasm, vocal palsy, and phono-trauma diseases. Computer aided diagnosis by analyzing human voice can be a remote and cost-effective tool for patients around the world. In this paper, we propose a deep transfer learning approach to differentiate pathological voice samples from normal ones. We utilize voice samples recorded from 200 patients at the Far Eastern Memorial Hospital (FEMH) to develop the deep transfer learning model. We extract prosodic, vocal tract and excitation features as new representations from the voice samples for diagnosis. To address the small data set challenge, we utilize the TIMIT dataset and develop a transfer learning approach in which a deep belief network (DBN) is first trained with the TIMIT data set. The trained model is then applied to the FEMH data set as a feature extractor. Finally, we train a support vector machine (SVM) classifier with the extracted features for diagnosis. We evaluate our approach using the leave one out cross validation (LOOCV) strategy on the 200 training patients, and achieve 94.90% sensitivity with 59.77% un-weighted average recall (UAR) for the 400 FEMH testing patients. Our results prove that the proposed method may be used effectively for pathological voice detection.

Development of a city cultivates regions with different functions such as working areas and entertainment venues. People in a city usually travel among these regions in certain movement patterns. Identifying those regions will facilitate government management and promote further development of the city. In this paper, we proposed a framework to identify urban functional regions in Chengdu city based upon mobility pattern and point of interest (POIs) information extracted from mobile check-ins data. Firstly, unlike GPS trajectories, location check-ins were discontinuous. Thus, the typical mobility patterns of location check-ins was mined. Secondly, an arrival/departure matrix based on the typical mobility patterns was constructed to obtain the topics of regions by clustering POIs. Because we considered a region's function as our topics, we transferred the problem into a topic modeling problem, and applied an improved probabilistic topic model to infer functions of the regions. We evaluated our approach with 227,428 check-ins in Chengdu collected from Sina Weibo from April 12 2012 to February 16 2013. The results showed that our method outperformed baseline methods solely clustering POIs.

Now, the urbanization process of China is accelerating. Urban land price is of great interest for the government to make reasonable policies and keep the healthy development of land market. Based on the data source of dynamic monitoring system and the statistical yearbook of Chengdu city, we identifies the related factors influencing the comprehensive land price of Chengdu. Firstly, we identified the nine strongly correlative factors of land price of Chengdu city. Secondly, we derived the land price for prediction. Thirdly, we compared the predicted land price with the real land price in the period of 2014–2015. Finally, the comprehensive land price of Chengdu in the period of 2017–2018 was forecasted with random forests and neural network, respectively. According to the results, we found that the error of the random forests is much smaller than that of neural network. Thus, we utilized random forests to predict the comprehensive land price of Chengdu in the period of 2017–2018. Our results showed that the comprehensive land price of Chengdu in the next two years would be stable and rises slightly.

In order to solve the problem of risk assessment for mountainous geological disaster in southwest of China, we selected Wenchuan county as the study area, where the geological disasters happen frequently. the digital elevation model (DEM), and other geographic data of Wenchuan county were utilized to evaluate the risk of geological disasters. Firstly, the weights of factors for geological hazard susceptibility were identified using analytic hierarchy process (AHP). Secondly, the fuzzy distinguish matrix based on the strength of membership function was established by combining AHP with fuzzy comprehensive evaluation (FCE). Thirdly, the geological hazard risk system was constructed according to the elevation, slope, distance from the river or the fault zone. Finally, we divided the study area into three risk types: high, moderate, and low. Our research results showed that the landslide numbers of high, moderate, and low risk are 11759, 16889, and 6075, respectively, and the corresponding percentages of area in Wenchuan county are 34%, 49%, 17%, respectively. Our results were in line with the historical disaster data. Therefore, the governments should pay more attentions to the geological disasters of these towns.

In the field of remote sensing, data acquired from a single sensor usually can't meet the needs of some special applications, because the information extracted from the data are often incomplete and limited. Data fusing can solve this problem, but it will lead to the redundant information. In this paper, we proposed a novel manifold learning approach to perform dimensionality reduction for the fusing optical and SAR data. And three typical manifold learning models, namely, ISOMAP, local linear embedding (LLE) and principle component analysis (PCA), were utilized to test the robustness of our method by comparing with the land cover classification results. Our experimental results showed that our proposed method obtained the best land cover classification results among these approaches for the fusing optical and SAR data.

Drought is a weather phenomenon widespread worldwide due to the water shortage or unbalance of supply and demand, and it's also one of the most serious natural disasters for human life and agricultural production. The middle reach of Yangtze river, one of China's most important grain producer, subjected to the sub-tropical monsoon climate, is prone to have droughts. This paper has practical implications as it build a model by depending on the normalized difference vegetation index (NDVI) and land surface temperature (LST) of moderate resolution imaging spectroradiometer (MODIS) between 2005 and 2009. Firstly, the 8-day LST and 16-day NDVI data, 8-day LST and 30-day NDVI data were utilized to construct the LST/NDVI feature space. Secondly, the temperature vegetation dryness index (TVDI) images of the reach were derived respectively. Thirdly, the temporal evolution and spatial variation of drought was analyzed. Finally, the results of two different years were compared to analyze the drought in the reach. Our study showed the drought in May was more severe than that in other months. Therefore, a severe drought event is more likely to happen in May in the middle reach of Yangtze river and more measures should be taken to alleviate the loss for the governments.

In this paper, we report a newfound vulnerability on smartphones due to the malicious use of unsupervised sensor data. We demonstrate that an attacker can train deep Convolutional Neural Networks (CNN) by using magnetometer or orientation data to effectively infer the Apps and their usage information on a smartphone with an accuracy of over 80%. Furthermore, we show that such attacks can become even worse if sophisticated attackers exploit motion sensors to cluster the magnetometer or orientation data, improving the accuracy to as high as 98%. To mitigate such attacks, we propose a noise injection scheme that can effectively reduce the App sniffing accuracy to only 15% and at the same time has negligible effect on benign Apps.

Post-traumatic stress disorder (PTSD) is a traumatic-stressor related disorder developed by exposure to a traumatic or adverse environmental event that caused serious harm or injury. Structured interview is the only widely accepted clinical practice for PTSD diagnosis but suffers from several limitations including the stigma associated with the disease. Diagnosis of PTSD patients by analyzing speech signals has been investigated as an alternative since recent years, where speech signals are processed to extract frequency features and these features are then fed into a classification model for PTSD diagnosis. In this paper, we developed a deep belief network (DBN) model combined with a transfer learning (TL) strategy for PTSD diagnosis. We computed three categories of speech features and utilized the DBN model to fuse these features. The TL strategy was utilized to transfer knowledge learned from a large speech recognition database, TIMIT, for PTSD detection where PTSD patient data is difficult to collect. We evaluated the proposed methods on two PTSD speech databases, each of which consists of audio recordings from 26 patients. We compared the proposed methods with other popular methods and showed that the state-of-the-art support vector machine (SVM) classifier only achieved an accuracy of 57.68%, and TL strategy boosted the performance of the DBN from 61.53% to 74.99%. Altogether, our method provides a pragmatic and promising tool for PTSD diagnosis.

Hyperspectral image (HSI) is usually composed of hundreds of bands which contain very rich spatial and spectral information. However, the high-dimensional data may lead to the curse of dimensionality phenomenon when it is used for land use classification or other applications, making it difficult to be utilized effectively. In this paper, we developed a deep learning classification framework based on the spectral and spatial information of hyperspectral image. Firstly, the deep learning features in different layers could be extracted automatically. Secondly, based on the learned deep learning features, we could obtain the classification of hyperspectral image with logistic regression (LR) classifier. Finally, we compared our approach with other methods including quadratic discriminant analysis with the multilevel logistic spatial prior (QDAMLL), logistic discriminant analysis with the multilevel logistic spatial prior (logDAMLL), linear discriminant analysis with the multilevel logistic spatial prior (LDAMLL), subspace multiclass logistic regression with the multilevel logistic spatial prior (MLRsub MLL), support vector machine on extended morphological profiles (SVM/EMP), support vector machine on expectation maximization and post-regularization (SVM-EM-PR). The experimental results showed that our method obtained the optimum accuracy, which was better than the other six approaches. And the OA was up to 99.39%. Therefore, the deep convolutional neural networks (DCNNs) is a robust method for land use classification with hyperspectral image.

Task engagement is delined as loadings on energetic arousal (affect), task motivation, and concentration (cognition) [1]. It is usually challenging and expensive to label cognitive state data, and traditional computational models trained with limited label information for engagement assessment do not perform well because of overlitting. In this paper, we proposed two deep models (i.e., a deep classilier and a deep autoencoder) for engagement assessment with scarce label information. We recruited 15 pilots to conduct a 4-h flight simulation from Seattle to Chicago and recorded their electroencephalograph (EEG) signals during the simulation. Experts carefully examined the EEG signals and labeled 20 min of the EEG data for each pilot. The EEG signals were preprocessed and power spectral features were extracted. The deep models were pretrained by the unlabeled data and were line-tuned by a different proportion of the labeled data (top 1%, 3%, 5%, 10%, 15%, and 20%) to learn new representations for engagement assessment. The models were then tested on the remaining labeled data. We compared performances of the new data representations with the original EEG features for engagement assessment. Experimental results show that the representations learned by the deep models yielded better accuracies for the six scenarios (77.09%, 80.45%, 83.32%, 85.74%, 85.78%, and 86.52%), based on different proportions of the labeled data for training, as compared with the corresponding accuracies (62.73%, 67.19%, 73.38%, 79.18%, 81.47%, and 84.92%) achieved by the original EEG features. Deep models are effective for engagement assessment especially when less label information was used for training.

Land use and land cover change (LUCC) is necessary to explore the factors leading to heavy drought and rainy-flood disaster in some districts of Sichuan province. A method based RS, GIS, GPS and Google earth (GE) is presented to establish LUCC database in Sichuan province and Chengdu district. At first, LUCC is interpreted based on the new temporal images and the land use and land cover database from TM in 2000.Secondly, some ground objects, which could not be identified in the new temporal images, were interpreted utilizing GE with some higher spatial resolution images. Thirdly, the new interpreted LUCC was validated in the field with GPS handheld receiver. Then, LUCC of Sichuan province was updated. A comparative analysis of LUCC between in Sichuan province and in Chengdu district was conducted and the result showed: (1) a large amount of farmland in Sichuan Province was occupied from 2000 to 2005 and the area is 84 573 ha. While construction land gained obviously and the area was 35 828 ha. The dynamic degree of construction land was 111.100/00from 2000 to 2005. The LUCC demonstrated that the economy of Sichuan province continued to develop, the cities were overspreading and the urban heat island effect was deteriorated from 2000 to 2005. (2) A large amount of farmland was also occupied in Chengdu district from 2000 to 2005, the area amounted to 12 989 ha. The farmland lost was mainly changed to construction land, amounting to 93%. And the dynamic degree was 117.410/00from 2000 to 2005, which was bigger than that in Sichuan province.

Hyperspectral remote sensing image (HSI) consists of hundreds of bands that contain rich space, radiation and spectral information. The high-dimensional data can also lead to the curse of dimensionality problem making it difficult to be used effectively. In this paper, we proposed a manifold learning algorithm to reduce the dimensionality for HSI data. For high dimensional datasets with continuous variables, it is often the case that the data points are arranged along with low dimensional structures, named manifolds, in the high dimensional space. Manifold learning aims to identifying those special low dimensional structures for subsequent usage such as classification or regression. However, many manifold learning algorithms perform an eigenvector analysis on a data similarity matrix whose size is N×N, where N is the number of data points. The memory complexity of the analysis is at least O(N2) that is not feasible for a regular computer to compute or storage for very large datasets. To solve this problem, we used statistical sampling methods to sample a subset of data points as landmarks. A skeleton of the manifold was then identified based on the landmarks. The remaining data points were then inserted into the skeleton by Locally Linear Embedding (LLE). We tested our algorithm on AVIRIS Salinas-A data set. The experimental results showed that the HSI dataset could be reduced to a lower-dimensional space for land use classification with good performance, and the main structure was preserved well.

High-dimensional data such as hyperspectral images contain abundant information of surface radiation. But the massive redundant information makes it complex to be utilized conveniently. To solve this problem, a manifold learning dimensionality reduction framework for hyperspectral image is proposed. Firstly, statistical sampling methods were used to sample a subset of data points as landmarks. A skeleton of the manifold was then identified basing on the landmarks. The remaining data points were then inserted into the skeleton by Locally Linear Embedding algorithm. At last, original data sets and data sets reduced with different manifold learning approaches were classified by KNN classifier to evaluate the performance of the proposed framework. The framework was tested on AVIRIS Salinas-A dataset. The experimental results showed that the tradeoff of accuracy with different landmarks is of great significant. Insufficient landmarks lead to low accuracy and excess landmarks may spend a considerable amount of time.

Under the indoor building environment, when the fires and other accidents occur, how to effectively organize the masses evacuation and fire rescue, is closely related to the safety of people's lives and property and has become a critical problem of public concern. This paper presents an improved ant colony algorithm (ACO) to solve the problem of how to optimize the evacuation route and rescue route when an accident occurs. According to the key factors affecting people emergency evacuation, such as indoor building environment, fire and its combustion products, problem of path's optimal selection, etc., we propose an emergency evacuation model, based on the model it can give an optimal evacuation route for the mass and an optimal rescue route for the firefighters. We also analyzes the search results, it shows that the search results is robust and reasonable.

In China, drought is one of the major environmental disasters, which bring great harm to the people. The middle reach of Yangtze River is the most important base to produce grains in China. Influenced by the summer monsoon, the drought occurs frequently. In our paper, the NDVI and LST from MODIS data were utilized to calculate the TVDI (Temperature Vegetation Dryness Index), which were used to monitor the drought of the study area. Meteorological drought indices were calculated from 10-day precipitation, temperature and evaporation data of 94 meteorological stations, including precipitation standardized variables, dryness and relative moisture index were used to analyze the degree of drought and the area of drought. The results showed that TVDI is significantly related to soil moisture.

A new, fast and computationally efficient lateral subpixel shift registration algorithm is presented. It is limited to register images that differ by small subpixel shifts otherwise its performance degrades. This algorithm significantly improves the performance of the single-step discrete Fourier transform approach proposed by Guizar-Sicairos and can be applied efficiently on large dimension images. It reduces the dimension of Fourier transform of the cross correlation matrix and reduces the discrete Fourier transform (DFT) matrix multiplications to speed up the registration process. Simulations show that our algorithm reduces computation time and memory requirements without sacricing the accuracy associated with the usual FFT approach accuracy.

Accurate classification of Alzheimer's disease (AD) and its prodromal stage, mild cognitive impairment (MCI), plays a critical role in possibly preventing progression of memory impairment and improving quality of life for AD patients. Among many research tasks, it is of a particular interest to identify noninvasive imaging biomarkers for AD diagnosis. In this paper, we present a robust deep learning system to identify different progression stages of AD patients based on MRI and PET scans. We utilized the dropout technique to improve classical deep learning by preventing its weight coadaptation, which is a typical cause of overfitting in deep learning. In addition, we incorporated stability selection, an adaptive learning factor, and a multitask learning strategy into the deep learning framework. We applied the proposed method to the ADNI dataset, and conducted experiments for AD and MCI conversion diagnosis. Experimental results showed that the dropout technique is very effective in AD diagnosis, improving the classification accuracies by 5.9% on average as compared to the classical deep learning methods.

A hierarchical elastic computer-aided detection algorithm is proposed to automatically detect the horizon in an aerial image. A hierarchical strategy, including coarse-level detection and fine-level adjustment, is applied. First, the original image is blurred by a large-scale low-pass filter. Then, a Canny edge detector and Hough transform are successively utilized to find major edges in the image and identify lines associated with those major edges. The desired horizon is modeled by the resulting line that best satisfies certain criteria. By doing so, the general position of the horizon can be quickly detected at the coarse-level step. Since the horizon is often not a straight line, an elastic fine-level adjustment is applied to capture the precise curvature of the horizon. A quantitative performance metric is designed, and preliminary experimental results show the feasibility and reliability of the proposed algorithm.

We present a three-step method to predict Prostate cancer (PCa) regions on biopsy tissue samples based on high confidence, low resolution PCa regions marked by a pathologist. First, we apply a texture analysis technique on a high magnification optical image to predict PCa regions on an adjacent tissue slice. Second, we design a prediction model for the same purpose using matrix-assisted laser desorption/ionization mass spectrometry (MALDI-MS) tissue imaging data from the adjacent slice. Finally, we fuse those two results to obtain the PCa regions that will assist MALDI imaging biomarker identification. Experiment results show that the texture analysis based prediction is sensitive (sen. 87.45%) but not specific (spe. 75%), and the prediction based on the MALDI spectra data is specific (spe. 100%) but less sensitive (sen. 50.98%). By combining those two results, a much better prediction for PCa regions on the adjacent slice can be achieved (sen. 80.39%, spe. 93.09%).

CT colonography (CTC) is an emerging noninvasive technique for screening and diagnosing colon cancers. Computer aided detection (CAD) techniques can increase sensitivity and reduce false positives. We propose to employ topographical height maps in our CAD pipeline. For every detection, a height map is computed using a ray-casting algorithm. Since colonic polyps are protrusions outward from the colon wall and are round in contour, their height maps present concentric patterns. The projection direction is optimized through a multi-scale spherical search. We derive several topographic features from the map, and also compute texture features from the Haar wavelet coefficients. We send the selected features to a committee of support vector machines for classification. We have tested our method on 1186 patients with 226 polyps. Results showed that the height map features can reduce false positives by about 50%.

A piecewise linear network is discussed which classifies N-dimensional input vectors. The network uses a distance measure to assign incoming input vectors to an appropriate cluster. Each cluster has a linear classifier for generating class discriminants. A training algorithm is described for generating the clusters and discriminants. Theorems are given which relate the network's performance to that of nearest neighbor and k-nearest neighbor classifiers. It is shown that the error approaches Bayes error as the number of clusters and patterns per cluster approach infinity.

We present an efficient feature selection algorithm for the general regression problem, which utilizes a piecewise linear orthonormal least squares (OLS) procedure. The algorithm 1) determines an appropriate piecewise linear network (PLN) model for the given data set, 2) applies the OLS procedure to the PLN model, and 3) searches for useful feature subsets using a floating search algorithm. The floating search prevents the "nesting effect." The proposed algorithm is computationally very efficient because only one data pass is required. Several examples are given to demonstrate the effectiveness of the proposed algorithm.

In the neural network literature, many preprocessing techniques, such as feature de-correlation, input unbiasing and normalization, are suggested to accelerate multilayer perceptron training. In this paper, we show that a network trained with an original data set and one trained with a linear transformation of the original data will go through the same training dynamics, as long as they start from equivalent states. Thus preprocessing techniques may not be helpful and are merely equivalent to using a different weight set to initialize the network. Theoretical analyses of such preprocessing approaches are given for conjugate gradient, back propagation and the Newton method. In addition, an efficient Newton-like training algorithm is proposed for hidden layer training. Experiments on various data sets confirm the theoretical analyses and verify the improvement of the new algorithm.

We study the demodulation problems for enhanced general packet radio services (EGPRS) wireless systems, where Rician flat fading channels are considered. A linear interpolation with decision feedback combined with a modified self-organizing-map (LIDF-MSOM) demodulator is implemented for such systems. Simulation results show that the performance of the proposed demodulator is much better than that of LIDF alone.

We study the demodulation problem in time division multiple access (TDMA) wireless asynchronous transfer mode (ATM) networks, where Rician flat fading channels are considered. A linear interpolation with decision feedback combined with a modified version of the self-organizing-map (LIDF-SOM) demodulator is proposed for such a system. We obtain the training sequence by exploiting medium access control (MAC) and data link control (DLC) protocols such that a semi-blind adaptive demodulator is implemented. Simulation results show that LIDF-SOM obtains 0.4-1.0 dB gain over Rician fading channels as compared to LIDF alone.

This paper studies the co-channel interference (CCI) problem for time-division-multiple-access (TDMA) cellular mobile communication systems with burst transmission. We present a method using learn vector quantization (LVQ) to cancel CCI for such systems. The model of the overall CCI is significantly simplified based on that of the individual CCI. The LVQ is realized as a classification equalizer with a decision feedback adaptive filter. An extremely small number of unique words (UWs) is utilized to initialize the LVQ equalizer. Simulation results show that the bit error rate (BER) of our proposed method is much better than that of the recently proposed nearest neighbor classification (NNC) equalizer.

Channel equalization has been revealed to be a classification problem by some recent applications of clustering and neural network techniques. In this paper, a new unsupervised learning (clustering) algorithm, adaptive nearest neighbor classifier (ANNC) is presented for channel equalization. ANNC can mine more channel characteristics that the recently proposed nearest neighbor (NNC) classifier. The proposed method is applied to a time-division-multiple-access (TDMA) satellite communication system with burst digital transmission. The improvement of the proposed algorithm over the recently reported NNC approach is clearly demonstrated.

In this paper, we study the demodulation problem in time-division-multiple-access (TDMA) wireless asynchronous transfer mode (ATM) networks. We propose a self-organizing-map (SOM) based demodulator. A linear interpolation with decision feedback (LIDF) combined with SOM algorithm (LIDF-SOM) is developed, in which the SOM network is initialized with some known symbols. Such known symbols are obtained by exploiting the medium access control (MAC) and data link control (DLC) protocols. Our scheme has three advantages, it: (1) avoids training symbols for uplink data bursts, (2) has reduced computational costs compared with common blind demodulators, and (3) blocks propagation estimation error through cells. Simulation results show that LIDF-SOM has better performances (0.2-0.4 dB gain) over time-varying channel as compared to the LIDF algorithm.