Effect of grouping in vector recognition system based on SOM

This chapter focuses on significant applications of self-organizing maps (SOMs), that is, unsupervised learning neural networks in two supply chain applications: cellular manufacturing and real-time management of a delayed delivery vehicle. Both problems require drastic complexity reduction, which is addressed effectively by clustering using SOMs. In the first problem, we cluster machines into cells and we use Latent Semantic Indexing for effective training of the network. In the second problem, we group the distribution sites into clusters based on their geographical location. The available vehicle time is distributed to each cluster by solving an appropriate non-linear optimization problem. Within each cluster an established orienteering heuristic is used to determine the clients to be served and the vehicle route. Extensive experimental results indicate that in terms of solution quality, our approach in general outperforms previously proposed methods. Furthermore, the proposed techniques are more efficient, especially in cases involving large numbers of data points. Neural networks have and will continue to play a significant role in solving effectively complex problems in supply chain applications, some of which are also highlighted in this chapter.

This paper describes a two-stage recognition method that reduces the calculation load of correlation and improves recognition accuracy in statistical image recognition. It consists of an image screening and recognition stage. Image screening selects a candidate set of subimages that are similar to the object class using a lower dimensional feature vector. Since recognition is made for the selected subimages set using a higher dimensional feature vector, overall recognition efficiency is improved. The classifier in recognition designed from the selected subimages also improves recognition accuracy because selected subimages are less contaminated than the original ones. A screening criterion for measuring overall efficiency and accuracy of recognition is introduced to be exploited in designing the feature spaces of image screening and recognition. The results of experiments for the eye- and mouth-area detection in face images and text-area detection in document images show that the designed feature spaces improve recognition accuracy and more efficiency than does the conventional one-stage recognition method.

<p><span>Understanding the drivers of present-day sea-level change is vital for improving sea-level projections and for adaptation and mitigation plans against sea-level rise. Sea-level budget (SLB) studies focus on attributing the observed sea-level change to its different drivers (steric and barystatic changes). While the global mean SLB is closed, explaining the drivers of sea-level change on a finer spatial scale leads to large discrepancies. Recent studies have shown that closing the regional budget on a regular 1x1˚ grid is not possible due to limitations of the observations itself, but also due to the spatial patterns and variability of the underlying processes. Consequently, the regional budget has been mainly analyzed on a basin-wide scale. </span></p><p><span> </span><span>In this study we use Self-Organizing Maps (SOM), an unsupervised learning neural network, to extract regions of coherent sea-level variability based on 27 years of satellite altimetry data. The SOM clusters have a higher level of spatial detail compared to entire ocean basins, while being large enough to allow for a consistent sea-level budget analysis. The clusters also show how sea-level variability is interconnected among different ocean regions (for example, due to large-scale climate patterns). We perform the clustering analysis on the Atlantic and Indo-Pacific Oceans separately, obtaining a total of 18 clusters. Preliminary results show that we can close the sea-level budget from 1993-2017 in 67% of the clusters. The regions with discrepancies highlight important regional processes that are affecting sea-level change and have not thus far been included in the sea-level budget. In this way, using neural networks provides new insight into regional sea-level variability and its drivers. </span></p>

The article develops multi-pattern formation exchange using A* searching algorithm, and programs the shortest motion paths for mobile robots. The system contains an image recognition system, a motion platform, some wireless RF modules and five mobile robots. We use Otsu algorithm to recognize the variety 2D bar code to classify variety pattern, and control five mobile robots to execute formation exchange, and present the movement scenario on the motion platform. We have been developed some pattern formations according to game applications, such as hook pattern formation, T pattern formation, L pattern formation, rectangle pattern formation, sward pattern formation and so on, and develop the user interface of the multi-robot system to program motion paths for variety pattern formation exchange on the supervised computer. The supervised computer programs pattern formation exchange according to the image recognition results, and controls mobile robots moving on the motion platform via wireless RF interface. In the experimental results, mobile robots can receive the pattern formation command from the supervised computer, and change the original pattern formation to the assigned pattern formation on the motion platform, and avoid other mobile robots on real-time.

A new image recognition system based on multiple linear regression is proposed. Particularly, there are a number of innovations in image segmentation and recognition system. In image segmentation, an improved histogram segmentation method which can calculate threshold automatically and accurately is proposed. Meanwhile, the regional growth method and true color image processing are combined with this system to improve the accuracy and intelligence. While creating the recognition system, multiple linear regression and image feature extraction are utilized. After evaluating the results of different image training libraries, the system is proved to have effective image recognition ability, high precision, and reliability.

Mobile Augmented Reality (AR), which overlays digital content on the real-world scenes surrounding a user, is bringing immersive interactive experiences where the real and virtual worlds are tightly coupled. To enable seamless and precise AR experiences, an image recognition system that can accurately recognize the object in the camera view with low system latency is required. However, due to the pervasiveness and severity of image distortions, an effective and robust image recognition solution for “in the wild” mobile AR is still elusive. In this article, we present CollabAR, an edge-assisted system that provides distortion-tolerant image recognition for mobile AR with imperceptible system latency . CollabAR incorporates both distortion-tolerant and collaborative image recognition modules in its design. The former enables distortion-adaptive image recognition to improve the robustness against image distortions, while the latter exploits the spatial-temporal correlation among mobile AR users to improve recognition accuracy. Moreover, as it is difficult to collect a large-scale image distortion dataset, we propose a Cycle-Consistent Generative Adversarial Network-based data augmentation method to synthesize realistic image distortion. Our evaluation demonstrates that CollabAR achieves over 85% recognition accuracy for “in the wild” images with severe distortions, while reducing the end-to-end system latency to as low as 18.2 ms.

A new, efficient clustering method for solving the cellular manufacturing problem is presented in this paper. The method uses the part-machine incidence matrix of the manufacturing system to form machine cells, each of which processes a family of parts. By doing so, the system is decomposed into smaller semi-independent subsystems that are managed more effectively improving overall performance. The proposed method uses Self Organizing Maps (SOMs), a class of unsupervised learning neural networks, to perform direct clustering of machines into cells, without first resorting to grouping parts into families as done by previous approaches. In addition, Latent Semantic Indexing (LSI) is employed to significantly reduce the complexity of the problem resulting in more effective training of the network, significantly improved computational efficiency, and, in many cases, improved solution quality. The robustness of the method and its computational efficiency has been investigated with respect to the dimension of the problem and the degree of dimensionality reduction. The effectiveness of grouping has been evaluated by comparing the results obtained with those of the k-means classical clustering algorithm.

The global anime market continues to grow, character recognition is particularly important in the derivatives market. And the demand for anime content on various platforms continues to increase. However, relatively few studies have been conducted on anime character image recognition, and the current research focuses on image recognition using a single ResNet50 model, with few studies attempting to incorporate other models to improve recognition accuracy. This project aims to construct a Pokémon image recognition system using deep learning techniques to improve classification efficiency and accuracy. The project uses the Pokémon image dataset on kaggle and performs data preprocessing, including data balancing, de-noising, normalization and data enhancement. The project constructed a classification prediction model combining ResNet50 and SVM respectively, which achieved 97.25% accuracy, 97.11% recall and 97.15% F1 score on the test set, indicating that the model can effectively recognize different kinds of Pokémon images. This project verifies the effectiveness of the ResNet50 model in image recognition tasks and provides a useful reference for subsequent research on image recognition.

Support vector machine (SVM) is one of the state-of-the-art tools for linear and nonlinear pattern classification. One of the design issues in SVM classifier is reducing the number of support vectors without compromising the classification accuracy. A technique denoted as diminishing learning (DL) is already proposed in literature for an SVM based multi-class isolated digit recognition system using speaker dependent TI46 database of isolated digits. In this paper, the computational complexity for SVM and SVM-DL based isolated digit recognition system is studied and the computation time for both the classifiers is evaluated by system-on-programmable-chip (SOPC) implementation of the recognition system onto an Altera Cyclone II Series FPGA using Nios II Soft-core processor. The number of support vectors is reduced by 38.28–90.25 % on using SVM-DL for isolated digit recognition problem. This in turn reduces the classification time for SVM-DL by 31.45–91.78 % over SVM. Recognition accuracies of 97 and 98 % are achieved for SVM classifier with and without DL technique, respectively. The study confirms the effect of, the order in which the classes are classified, on the recognition accuracy. For the TI46 database, about 2–4 % increase in recognition accuracy is obtained by choosing the optimum order for SVM-DL classifier. The proposed SOPC implementation of SVM-DL based recognition system can be employed for various other pattern recognition applications too such as face recognition, character recognition and target recognition.

Clustering methods aims to organize a set of items into clusters such that items within a given cluster have a high degree of similarity, while items belonging to different clusters have a high degree of dissimilarity. The self-organizing map (SOM) introduced by Kohonen is an unsupervised competitive learning neural network method which has both clustering and visualization properties, using a neighborhood lateral interaction function to discover the topological structure hidden in the data set. In this paper, we introduce a batch self-organizing map algorithm based on adaptive distances. Experimental results obtained in real benchmark datasets show the effectiveness of our approach in comparison with traditional batch self-organizing map algorithms.

With the growth of society, people have higher and higher requirements for the quality of life, and the Internet has become an indispensable part of our daily life. Ants are very typical, common, convenient, creative and convenient. This paper mainly introduces the methods based on the Internet of Things and ant colony computing. By analyzing the research status of ant algorithm at home and abroad and relevant literature, we draw conclusions, and propose improvement plans to improve the theoretical system in this field, further optimize the image recognition application of medical equipments and medicine materials in the Internet of Things environment. Then, according to the system functions to be achieved in this paper, we determine the objective function, design indicators and parameters, so as to extract features. Finally, we get the optimal solution of the feature vector, and then send the data to the background database to obtain the recognition results, And verify the model in the experiment This paper designs a simple, low-cost, efficient and high-precision recognition system based on the ant algorithm to test. Through the image recognition experiment for medical equipments and drug materials in the Internet of Things environment, the image recognition system based on the ant algorithm achieves the feature extraction time within 20 seconds, while driving the system recognition time to reach 26 seconds, with a feature matching rate of more than 82%, which can fully meet the user’s image recognition needs, The scheme not only saves resources, but also has high practical value.

We have been working on the application of Machining Learning in Metabolomics, Drug-Drug Interaction Discovery and Human Gait Recognition [1-5], profiling large data sets. Extraction of vital information about 1) plant metabolomics that can improve the environment and food quality, 2) in vitro neuronal network behavior patterns for various drugs that can be used to characterize drugs for brain deceases and 3) human gait patterns that can reveal various diseases were among these applications. We have demonstrated with considerable success in using unsupervised clustering techniques to analyze genetic and metabolomic data. This includes analysis of drought resistance in wheat [4] and microbial metagenomes [5].We introduced two methods: Near Unsupervised Learning (NUL) and Sub-sample Error Graphs (SEGs) [5] to analyze large amount of data. Self Organizing Map (SOM), which is one of the widely used Unsupervised Neural Networks, has been used as a data-mining tool due to its ability to map high dimensional data into a two dimensional feature map, which is expected to be topology preserving allowing users to visually identify clusters by their topological relationships in terms of their proximity on the map. The Growing Self Organizing Maps (GSOM) further allows the map size to be determined by the algorithm, which relies upon a user set parameter called Spread Factor (SF) [6]. The wide availability of GPUs for affordable prices allows faster comparison of various SOMs with different maps sizes taking away some of the advantages GSOM claimed. Further development of GSOM into a Dynamic SOM Tree exploits the possibility of varying SF to obtain multiple GSOMs from a small number of compact clusters to a large number of sparse clusters [7]. NUL methods can be applied on GSOM using a small number of labels that should be available for every class. This is however not realistic in some applications where the number of classes cannot be explicitly known. We propose a new method call Deep Near Unsupervised Learning (D-NUL), where Dynamic SOM tree is used instead of GSOM and the number of classes are not assumed to be known. The implementation of Dynamic SOM tree methods with varying SF on GPUs will make the computation with D-NUL possible for many problems in big data analytics.

In order to improve the accuracy and efficiency of Internet of things illegal intrusion identification, an automatic identification method of Internet of things illegal intrusion based on unsupervised learning is proposed. Firstly, in order to accurately and automatically identify the illegal intrusion of the Internet of things, the gradient descent clustering method is used to fully collect the illegal intrusion data of the Internet of things. Secondly, the kernel principal component analysis method is used to extract the characteristics of online intrusion data. Finally, unsupervised learning neural network is used to train the above extracted intrusion data features to complete the automatic identification of illegal intrusion in the Internet of things. The experimental results show that compared with the traditional intrusion recognition methods, the recognition accuracy and efficiency of this method have been significantly improved.

Early detection of gastric cancer dramatically improves patient outcomes, yet visual diagnosis during endoscopy remains operator-dependent and time-consuming. This paper presents an end-to-end Image Recognition and Diagnosis System for Early Gastric Cancer (IRDS-EGC) that combines deep convolutional neural networks for lesion detection, segmentation and classification on endoscopic images. We propose a hybrid architecture that ensembles a classification backbone (ResNet50 and EfficientNet-B3) with a U-Net-based segmentation head to localize suspicious regions and then aggregate features for final diagnosis. The system was trained and validated on a curated dataset of 10,000 annotated endoscopic images (3,500 early-gastric-cancer images, 6,500 benign) with clinically-informed augmentation and pre-processing. On a held-out test set (n = 1,500), the proposed system achieved a sensitivity of 94.3%, specificity of 90.3%, accuracy of 91.7%, and area under the ROC curve (AUC) of 0.955 for the binary task of early cancer vs non-cancer. Qualitative analysis shows the segmentation output closely aligns with clinician annotations. We discuss clinical integration pathways, limitations, and directions for future work including prospective validation and real-time endoscopic deployment.

In order to achieve intelligent, fast and accurate grading of apples, Apple's image acquisition and recognition system has been built, which is based on machine vision technology. Through color-to-gray, image noise reduction, image enhancement, and image segmentation processing of the collected image, three characteristics of the shape, size and color of the apple were recognized. Taking the 3 features of the apple identified from the image recognized as input, a multi-feature fusion classification method based on image processing and SVM was proposed in this paper. Through the constructed SVM vector machine, the apple's shape, size and color are synthesized to judge the apple's grade. The experimental results show that the classification method has the characteristics of high recognition rate and good real-time performance. This method lays the foundation for the application of machine vision technology in apple grading.