Detecting surface defects plays a crucial role in ensuring the quality, functionality, and security of the production process. Traditional image processing techniques and machine learning models rely on manual analysis and feature extraction for specific vision inspection tasks. Deep learning approaches, which can automatically extract features from images, have demonstrated outstanding performance in computer vision tasks, including detecting surface defects. Motivated by this consideration, a Systematic Literature Review (SLR) method is employed for the comprehensive analysis of studies published between 2020 and 2023 in the field of deep learning-based surface defect detection applications in industrial products. The study provides a technical taxonomy for deep learning models according to the content of current studies through the SLR process, including Convolutional Neural Networks (CNN), encoder–decoder models, pyramid network models, Generative Adversarial Networks (GAN), attention-based models, and other models for surface defect detection. Then, the commonly used datasets for surface defect detection are discussed, and a comparative analysis of deep learning models’ performance is provided. Our comparative analysis reveals that pyramid network models and CNN models are the most frequently used deep learning models for surface defect detection. These models yield reasonable results in surface defect detection due to their exceptional feature extraction capabilities. Finally, some hints for addressing future research directions and identifying open issues in surface defect detection applications are presented.

This study aims to explore the visualization of relational structures in composition within visual imagery. Composition is an essential technique used by artists to manage the elements within an image, facilitating the communication and transmission of messages through the organization of visual elements. Human thinking involves a form of structural thinking, which is also triggered when appreciating works of art. This research seeks to incorporate the characteristics of human structural thinking into composition, making the relationships between compositions logical and computable. Ultimately, the study proposes a method to visualize the relationships within image compositions. The proposed method begins by inviting experts to explain the key knowledge within the artwork. This helps in understanding the content of the artwork and establishing nodes on the image to represent relevant compositional knowledge. These knowledge contents are then presented digitally and in a structured form. Next, the experiences of viewers are transformed into precise data. The gaze points generated while observing the image are connected. Finally, the data from both experts and viewers are combined to reveal the relational structures within the composition. This method helps overcome the limitations of previous understandings of artworks, which heavily relied on subjective interpretations by experts and superficial descriptions based solely on the visual organization of the image. This research aims to provide an objective perspective in interpreting composition by exploring the hidden structures organized within the image from the viewpoint of viewer attention. The research findings indicate that the compositions formed by viewers are not merely static or dynamic structures. Instead, they represent experiential, dynamic, and vital organizational structures. The results of the study can better approach the viewer's real visual experiences and confirm that composition is not just a technique but a purpose.

Most of real-world datasets are of mixed type including both numeric and categorical attributes. Unlike numbers, operations on categorical values are limited, and the degree of similarity between distinct values cannot be measured directly. In order to properly analyze mixed-type data, dedicated methods to handle categorical values in the datasets are needed. The limitation of most existing methods is lack of appropriate numeric representations of categorical values. Consequently, some of analysis algorithms cannot be applied. In this paper, we address this deficiency by transforming categorical values to their numeric representation so as to facilitate various analyses of mixed-type data. In particular, the proposed transformation method preserves semantics of categorical values with respect to the other values in the dataset, resulting in better performance on data analyses including classification and clustering. The proposed method is verified and compared with other methods on extensive real-world datasets.

The original self-organizing map (SOM) was proposed in the context of processing numeric data. In previous studies, an extended SOM incorporating data structure distance hierarchies has been proposed to facilitate handling of categorical values. The model could take into account the semantics embed ded in categorical values via distance hierarchies. In addition to manual construction by domain experts, an approach to learning distance hierarchies from datasets has been devised. However, the proposed approach in the previous study was based on supervised learning which demands presence of a class attribute in the dataset. In real-world applications, class attribute may not be available. Thus, the supervised approach can be inapplicable. In this article, we present several methods of unsupervised learning of distance hierarchies so that neither are class attribute nor domain experts required in measuring similarity degree between categorical values. We then integrate the learned distance hierarchies with the extended SOM to facilitate the application to datasets without a class attribute. We conduct experiments to verify feasibility and compare performance of the proposed unsupervised-learning methods.

Visualization is a useful technique in data analysis, especially, in the initial stage, data exploration. Since high-dimensional data is not visible, dimensionality reduction techniques are usually used to reduce the data to a lower dimension, say two, for visualization. In previous studies, dimens ionality reduction was investigated in the context of numeric datasets. Nevertheless, most of real-world datasets are of mixed-type containing both numeric and categorical attributes. In this case, a traditional approach could neither handle it directly nor output appropriate results. To address this problem, we propose a procedure for visualized analysis of mixed-type data via dimensionality reduction. Dissimilarity between categorical values is learned from the dataset and further used to measure the distance between mixed-type data points. In addition, we propose an approach to identifying significant features and visualizing patterns from the projection map chosen according to quality measures. Experiments on real-world datasets were conducted to demonstrate feasibility of the proposed method.

Many real-world datasets are of mixed types, having numeric and categorical attributes. Even though difficult, analyzing mixed-type datasets is important. In this paper, we propose an extended self-organizing map (SOM), called MixSOM, which utilizes a data structure distance hierarchy to facilitate the handling of numeric and categorical values in a direct, unified manner. Moreover, the extended model regularizes the prototype distance between neighboring neurons in proportion to their map distance so that structures of the clusters can be portrayed better on the map. Extensive experiments on several synthetic and real-world datasets are conducted to demonstrate the capability of the model and to compare MixSOM with several existing models including Kohonen's SOM, the generalized SOM and visualization-induced SOM. The results show that MixSOM is superior to the other models in reflecting the structure of the mixed-type data and facilitates further analysis of the data such as exploration at various levels of granularity.

Nowadays, a fault management incorporating with alarm correlation rules play an important role on network management and maintenance. Alarms are unsolicited messages received from network elements or network management systems carrying information about malfunction location, type, severity and other relevant information. Usually, alarm information consists of categorical and numerical attributes and is difficult to be processed by most of clustering algorithms. This paper introduces a method of mining correlation rules of alarm messages by modified adaptive resonance theory network (MART) algorithm. The experiments illustrate good performance of the proposed algorithm with concept hierarchy tree to solve similar degrees of mixed data. Furthermore, this method could be used to build a hybrid system with traditional case-based reasoning system.

Vital signs provide critical information about the health of a patient. There are four vital signs commonly used by healthcare professionals to evaluate a patient's health. These are body temperature, pulse rate, respiration rate (rate of breathing), and blood pressure. The artificial neuromolecular system previously constructed in our group is a biologically motivated system that captures the biological structure-function relationships, and that possesses several features that facilitate evolutionary learning. This paper describes the application of the ANM system to evaluate the health of patients in respiration intensive care unit (RICU) from their vital signs. Experimental results show that the ANM system achieves a satisfactory result in accessing and predicting the health of patients, based on their vital signs.

Organisms have better adaptability that computer systems in dealing with environmental changes or noise. A close structure-function relation inherent in biological structures is an important feature for providing great malleability to environmental changes. By contrast, computers have fast processing speeds but with limited adaptability. A biologically motivated model (hardware design) that combines intra-and inter-neuronal information processing implemented with digital circuit was proposed. Pattern recognition was the present application domain. The circuit was tested with Quartus II system, a digital circuit simulation tool. The experimental result showed that the artificial neuromolecularware (ANM) exhibited a close structure-function relationship, possessed several evolvability-enhancing features combined to facilitate evolutionary learning, and was capable of functioning continuously in the face of noise.

Self-Organization Map (SOM) offers an effective visualization capability for analyzing high-dimensional data. Nevertheless, most SOM models lack a robust solution to appropriately manipulate both numeric and categorical data. To solve the foregoing problem, Generalized SOM (GenSOM) was proposed to handle distance measurement of mixed-type data via distance hierarchy. Whereas GenSOM constrains projection result in a predetermined fixed-size map, making the resultant map unable to reflect data distribution in accordance with the nature of data clusters. In this paper, we propose a Growing Mixed-type SOM (GMixSOM) which extends GenSOM with a dynamic structure, to handle mixed-type data and tackle the problem of fixed map structure of GenSOM. Experimental results show the proposed method can reveal topological relationship between mixed data and overcome the drawback of map structure constraint arisen in GenSOM.

The World Wide Web has been considered one of the important sources for information. Using search engines to retrieve Web pages can gather lots of information, including foreign information. However, to be better understood by local readers, proper names in a foreign language, such as English, are often transliterated to a local language such as Chinese. Due to different translators and the lack of translation standard, translating foreign proper nouns may result in different transliterations and pose a notorious headache. In particular, it may cause incomplete search results. Using one transliteration as a query keyword will fail to retrieve the Web pages which use a different word as the transliteration. Consequently, important information may be missed. We present a framework for mining synonymous transliterations as many as possible from the Web for a given transliteration. The results can be used to construct a database of synonymous transliterations which can be utilized for query expansion so as to alleviate the incomplete search problem. Experimental results show that the proposed framework can effectively retrieve the set of snippets which may contain synonymous transliterations and then extract the target terms. Most of the extracted synonymous transliterations have higher rank of similarity to the input transliteration compared to other noise terms.

When editors of newspapers and magazines translate proper nouns from foreign languages into Chinese, the Chinese translation (termed transliterations) they choose will typically be phonetically similar to the original word. With many different translators working without a common standard, there may be many different Chinese transliterations for the same proper noun, such as using the same sounds but different Chinese characters or even using different sounds and characters. This causes confusion for the reader and, more importantly, leads to incomplete Chinese Web search results. This article investigates the similarity comparison of transliterations as a first step toward solving the incomplete search problem. We devise a method based on comparing digitalized Chinese character (or Hanzi) sounds. Along with four other methods based on comparing grapheme or phoneme similarity, we compare their performance of identifying synonymous transliterations against noise words taken from Web pages. Experimental results indicate that our method surpasses the other methods due to its advantage of containing more discriminative information in sound vectors. The method performing the second best is based on a scheme which assigns similarity between phonemes by carefully considering articulatory features of phonemes, including using multivalued features and placing different weights on the features. Among six pinyin schemes used to romanize Chinese transliterations, the Tongyong scheme outperforms the others.

Data mining has become a popular technology in analyzing complex data. Clustering is one of the data mining core techniques. In this paper, we propose a new visualized clustering approach, namely generalized visualization-induced self-organizing map (GViSOM), to cluster mixed, numeric and categorical, data. Our model integrates the ideas from SOM, GSOM, and ViSOM, and overcomes their shortcomings, including projection distortion on the maps and the incapability of handling mixed data. GViSOM can directly handle mixed data and preserves the topological structure of the original data as faithfully as possible. Experimental results of a synthetic and a real datasets demonstrate that GViSOM is able to cluster mixed data and better reveals the cluster structure than SOM, GSOM and ViSOM.

When newspapers and magazines translate proper nouns from foreign languages into Chinese, the Chinese translation (termed transliterations) they choose will typically be phonetically similar to the original word. With many different translators working without a common standard, there may be many Chinese transliterations for the same proper noun, such as using the same sounds but different Chinese characters, or even using different sounds and characters. This causes confusion for the reader and, more importantly, leads to incomplete Chinese web search results. This paper investigates the similarity comparison of transliterations as a first step toward solving the incomplete search problem. Our research framework had two stages: training and recognition. In the training stage, we compared Chinese speech sounds and constructed a database of similarity matrices. In the recognition stage, we first convert transliterations to phonetic notation and then apply the matrices from the database to calculate the degree of similarity among different transliterations. Highly similar transliterations are likely to be synonymous, referring to the same foreign proper noun. Our research results indicate that our methodology achieves better performance compared to traditional pinyin approaches.

Data mining uncovers hidden, previously unknown, and potentially useful information from large amounts of data. Compared to the traditional statistical and machine learning data analysis techniques, data mining emphasizes providing a convenient and complete environment for the data analysis. In this paper, we propose an integrated framework for visualized, exploratory data clustering, and pattern extraction from mixed data. We further discuss its implementation techniques: a generalized self-organizing map (GSOM) and an extended attribute-oriented induction (EAOI), which not only overcome the drawbacks of their original algorithms, but also provide additional analysis capabilities. Specifically, the GSOM facilitates the direct handling of mixed data, including categorical and numeric values. The EAOI enables exploration for major values hidden in the data and, in addition, offers an alternative for processing numeric attributes, instead of generalizing them. A prototype was developed for experiments with synthetic and real data sets, and comparison with those of the traditional approaches. The results confirmed the feasibility of the framework and the superiority of the extended techniques.

The self-organizing map (SOM) is an unsupervised neural network which projects high-dimensional data onto a low-dimensional grid and visually reveals the topological order of the original data. Self-organizing maps have been successfully applied to many fields, including engineering and business domains. However, the conventional SOM training algorithm handles only numeric data. Categorical data are usually converted to a set of binary data before training of an SOM takes place. If a simple transformation scheme is adopted, the similarity information embedded between categorical values may be lost. Consequently, the trained SOM is unable to reflect the correct topological order. This paper proposes a generalized self-organizing map model that offers an intuitive method of specifying the similarity between categorical values via distance hierarchies and, hence, enables the direct process of categorical values during training. In fact, distance hierarchy unifies the distance computation of both numeric and categorical values. The unification is done by mapping the values to distance hierarchies and then measuring the distance in the hierarchies. Experiments on synthetic and real datasets were conducted, and the results demonstrated the effectiveness of the generalized SOM model.