TDEC: Deep Embedded Image Clustering with Transformer and Distribution Information

Learning deep discriminative representations with pseudo supervision for image clustering

Rapid growth in numbers of connected devices including sensors, mobile, wearable, and other Internet of Things (IoT) devices, is creating an explosion of data that are moving across the network. To carry out machine learning (ML), IoT data are typically transferred to the cloud or another centralized system for storage and processing; however, this causes latencies and increases network traffic. Edge computing has the potential to remedy those issues by moving computation closer to the network edge and data sources. On the other hand, edge computing is limited in terms of computational power, and thus, is not well-suited for ML tasks. Consequently, this article aims to combine edge and cloud computing for IoT data analytics by taking advantage of edge nodes to reduce data transfer. In order to process data close to the source, sensors are grouped according to locations, and feature learning is performed on the close by edge node. For comparison reasons, similarity-based processing is also considered. Feature learning is carried out with deep learning - the encoder part of the trained autoencoder is placed on the edge and the decoder part is placed on the cloud. The evaluation was performed on the task of human activity recognition from sensor data. The results show that when sliding windows are used in the preparation step, data can be reduced on the edge up to 80% without significant loss in accuracy.

A deep autoencoder feature learning method for process pattern recognition

Scene classification plays an important role in the interpretation of remotely sensed high-resolution imagery. However, the performance of scene classification strongly relies on the discriminative power of feature representation, which is generally hand-engineered and requires a huge amount of domain-expert knowledge as well as time-consuming hand tuning. Recently, unsupervised feature learning (UFL) provides an alternative way to automatically learn discriminative feature representation from images. However, the performances achieved by conventional UFL methods are not comparable to the state-of-the-art, mainly due to the neglect of locally substantial image structures. This paper presents an improved UFL algorithm based on spectral clustering, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">named</i> UFL-SC, which cannot only adaptively learn good local feature representations but also discover intrinsic structures of local image patches. In contrast to the standard UFL pipeline, UFL-SC first maps the original image patches into a low-dimensional and intrinsic feature space by linear manifold analysis techniques, and then learns a dictionary (e.g., using K-means clustering) on the patch manifold for feature encoding. To generate a feature representation for each local patch, an explicit parameterized feature encoding method, i.e., triangle encoding, is applied with the learned dictionary on the same patch manifold. The holistic feature representation of image scenes is finally obtained by building a bag-of-visual-words (BOW) model of the encoded local features. Experiments demonstrate that the proposed UFL-SC algorithm can extract efficient local features for image scenes and show comparable performance to the state-of-the-art approach on open scene classification benchmark.

Image processing has become a foundational component of modern intelligent systems, enabling automated interpretation and analysis of visual data across a wide range of application domains. However, traditional image processing techniques and classical machine learning models rely heavily on handcrafted features and rule-based representations, which significantly limit their adaptability, scalability, and performance when dealing with complex and high-dimensional image data. To overcome these limitations, this research paper presents an advanced machine learning–based image processing framework developed from an empirical dissertation study. The proposed framework employs a supervised deep learning approach using convolutional neural networks to perform end-to-end image classification. A systematic methodology involving image preprocessing, feature learning, CNN-based model development, training, validation, and comprehensive evaluation is adopted to ensure robustness and reliability. The experimental evaluation is conducted on a structured dataset consisting of 10,000 labeled images distributed across two classes. Model performance is assessed using standard classification metrics, including accuracy, precision, recall, F1-score, confusion matrix analysis, and training–validation learning curves. Experimental results demonstrate that the proposed model achieves an overall classification accuracy of 98.91 percent, with consistently high precision and recall values for both classes. Confusion matrix analysis reveals strong diagonal dominance with minimal misclassification, while learning curves confirm stable convergence and controlled overfitting. The findings highlight the effectiveness of advanced machine learning techniques in enhancing image processing accuracy, robustness, and scalability

Clustering plays important role in many areas such as medical applications, pattern recognition, image analysis and statistical data analysis. Image clustering is an application of image analysis in order to support high-level description of image content for image understanding where the goal is finding a mapping of the images into clusters. This paper presents an Artificial Bee Colony (ABC) based image clustering method to find clusters of an image where the number of clusters is specified. The proposed method is applied to three benchmark images and the performance of it is analysed by comparing the results of K-means and Particle Swarm Optimization (PSO) algorithms. The comprehensive results demonstrate both analytically and visually that ABC algorithm can be successfully applied to image clustering.

Knowing exact number of clusters in a digital image significantly facilitates in precisely clustering an image. This paper proposes a new technique for extracting exact number of clusters from grey scale images. It analyzes the contents of the input image and adaptively reserves one distinct cluster for one distinct grey value. The total count of the grey values found in an image determines the exact number of clusters. Based on the contents of image, this number of clusters keeps on changing from image to image. After obtaining this number, it is given as an input to Gaussian Mixture Model (GMM) which clusters the image.GMM works with finite number of clusters and forms mixture of various spectral densities contained in that image. The proposed method facilitates GMM to adapt itself according to the changing number of clusters. Therefore, the proposed model along with the inclusion of GMM, is named as Adaptive Finite Gaussian Mixture Model (AFGMM). The clustering performance of AFGMM is evaluated through Mean Squared Error (MSE) and Peak Signal to Noise Ratio (PSNR). Both of these performance measuring methods confirmed that exact number of clusters is essentially important for reliably analyzing an image.

Deep residual networks (ResNets) can learn deep feature representation from hyperspectral images (HSIs), and therefore have been widely used for HSI classification. Despite their high accuracies, there still exist a lot of challenging cases, such as open world recognition, limited-sample learning and visualization of learned classification features, which cannot be well addressed. Most of the challenges in HSI classification can be attributed to the dependence on softmax based loss function and classifiers, which cause the lack of robustness for deep learning models and the hardness to visualize the learned classification features. To improve the robustness and achieve the visualization of learned classification features, we propose a novel learning framework called Residual Prototype Learning Network, a combination of residual network and prototypes learning mechanism. Under the framework, a prototype learning based loss function is proposed to enhance intra-class compactness and the inter-class separation of these feature representations; in addition, a prototype learning based classifier is simultaneously proposed to achieve the 2D or 3D visualization of the classification features. The effectiveness of our proposed learning framework is evaluated on several publicly available HSI benchmarks, and the experimental results show that our approach achieve better results than traditional softmax based ResNets.

Intelligent colour analysis based on deep learning has a wide range of applications in image, video, and other colour information automatic analysis and understanding and has great potential. Deep learning models can learn the representation of color features through the process of feature abstraction and hierarchical extraction and feature learning to express colors, as well as pattern recognition that is suitable for the given task. They have stronger feature representation, learning, and generalization abilities. These characteristics allow deep learning models to better deal with the problems incurred by traditional methods in the processing, analysis, and understanding of colour and its features. The author mainly uses Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) to analyse and process color data through the designed experimental system, respectively. The learning process of deep learning model is fully automatic, and the representation and abstraction ability of the model is improved through the large-scale annotated data. It can be used in the field of image/video recognition and understanding of color by deep learning method. The experimental results are shown in this paper. The application of deep learning-based color intelligent analysis in different tasks has achieved remarkable results. A deep learning model can accurately analyse the color and color texture of the image and its distribution in the image, and be used in the image recognition and classification task.

Conventional machine learning methods use the selected hand-crafted features for activity recognition (AR). Motivated by the recent trends of AR applications in both human and robots using deep learning with automatically learned features, this paper explores to compare the performance of automatically-learned features by the deep networks and more comprehensive hand-crafted features with the conventional classification methods. We design and optimize the Recurrent Neural Networks (RNN) with the learned features and apply Support Vector Machine (SVM) and Random Forest (RF) with the completed hand-crafted features. We use two datasets for evaluation, i.e., a ground-truth dataset from human and a benchmark dataset from robots. The experimental results indicate that the learned features by deep networks and the hand-crafted features can perform equally well on the robot dataset; the hand-crafted features even perform better than the trained RNN networks on the ground-truth data.

Clustering is the unsupervised process of assigning entities into groups based on similarities among those entities. Image clustering is the crucial step of mining satellite images. As the satellite imagery is getting generated at a higher rate than the previous decades, it becomes essential to have better solutions in terms of accuracy as well as performance. In this paper, we are proposing the solution over big data platform Apache Spark which performs the clustering of images using different methods viz. Scalable K-means++, Bisecting K-means and Gaussian Mixture. Since the number of clusters is not known in advance in any of the methods, we also propose a Best of Breed approach of validating the number of clusters using Simple Silhouette Index algorithm and thus to provide the best clustering possible.

Malicious software (ransom ware) cyber attacks in frequency and severity, posing an increasingly serious threat to computer systems everywhere. Malware detection is a hot study area as several multiple computers, organisations, and governments have been affected by an exponential rise in malware attacks. Dynamic and static assessment of malicious characteristics and behaviour patterns is time expensive and useless in real-time malware detection, according to current technologies. It is becoming increasingly common for malicious apps to use polymorphic and adaptive techniques to rapidly modify their behaviour and develop a number of new malicious apps. In order to undertake an effective malware analysis, machine learning techniques (MLAs) are increasingly being used to create new malware varieties. This approach is time-consuming since it requires considerable feature engineering, learning and representation of features. Moreover the feature extraction process could be effectively eliminated by using advanced MLAs like deep learning. These methods have been shown to perform better with a biased training dataset, which restricts their practical application in real-time scenarios. A new improved approach for successful zero-day malware detection must be developed in order to eliminate biases and analyze these approaches autonomously.

This paper presents a comparison study between methods of deep learning as a new category of slope stability analysis, built upon the recent advances in artificial intelligence and conventional limit equilibrium analysis methods. For this purpose, computer code was developed to calculate the factor of safety (FS) using four limit equilibrium methods: Bishop’s simplified method, the Fellenius method, Janbu’s simplified method, and Janbu’s corrected method. The code was verified against Slide2 in RocScience. Subsequently, the average FS values were used to approximate the “true” FS of the slopes for labeling the images for deep learning. Using this code, a comprehensive dataset of slope images with wide ranges of geometries and soil properties was created. The average FS values were used to label the images for implementing two deep learning models: a multiclass classification and a regression model. After training, the deep learning models were used to predict the FS of an independent set of slope images. Finally, the performance of the models was compared to that of the conventional methods. This study found that deep learning methods can reach accuracies as high as 99.71% while improving computational efficiency by more than 18 times compared with conventional methods.

To address the problems in the existing video super-resolution methods, such as noise, over smooth and visual artifacts, which are caused by the reliance on limited external training or mismatch of internal similarity patch instances, this study proposes a novel video super-resolution reconstruction algorithm based on deep learning and spatio-temporal feature similarity (DLSS-VSR). The video super-resolution reconstruction mechanism with the joint internal and external constraints is established utilizing the complementary advantages of both external deep correlation mapping learning and internal spatio-temporal nonlocal self-similarity prior constraint. A deep learning model based on deep convolutional neural network is constructed to learn the nonlinear correlation mapping between low-resolution and high-resolution video frame patches. A novel spatio-temporal feature similarity calculation method is proposed, which considers both internal video spatio-temporal self-similarity and external clean nonlocal similarity. For the internal spatio-temporal feature self-similarity, we improve the accuracy and robustness of similarity matching by proposing a similarity measure strategy based on spatio-temporal moment feature similarity and structural similarity. The external nonlocal similarity prior constraint is learned by the patch group-based Gaussian mixture model. The time efficiency for spatio-temporal similarity matching is further improved based on saliency detection and region correlation judgment strategy, which achieves a better tradeoff between super-resolution accuracy and speed. Experimental results demonstrate that the DLSS-VSR algorithm achieves competitive super-resolution quality compared to other state-of-the-art algorithms in both subjective and objective evaluations.

As mobile devices and personal computers have been more frequently used through the Internet, data generated by not only people but also devices have been continuously piling up. The data growing endlessly is called Big Data and Deep Learning algorithms with the Big Data have been introducing the next level of artificial intelligence. It is generally applicable that the more data deep learning algorithms train, the more accurate the deep learning algorithms are. Then, an important problem is which size of data is enough for deep learning algorithms to train the data. In many cases, it is not practical that we wait for the data to grow bigger enough, and thus we need a new learning model that can reduce this latency time and timely derive learning results with useful accuracy. In this paper, we propose novel block-incremental learning models for deep learning and experimentally show that the proposed model can timely derive learning results with useful accuracy and the final accuracy is even better than the traditional deep learning algorithms with the same size of training data.