Robust Feature Extraction Research Articles

PBCS-ConvNeXt: Convolutional Network-Based Automatic Diagnosis of Non-alcoholic Fatty Liver in Abdominal Ultrasound Images.

Non-alcoholic fatty liver disease (NAFLD) is a highly prevalent chronic liver condition characterized by excessive hepatic fat accumulation. Early diagnosis is crucial as NAFLD can progress to more severe conditions like steatohepatitis, fibrosis, cirrhosis, and hepatocellular carcinoma without timely intervention. While liver biopsy remains the gold standard for NAFLD assessment, abdominal ultrasound (US) imaging has emerged as a widelyadopted non-invasive modality due to convenience and low cost. However, the subjective interpretation of US images is challenging and unpredictable. This study proposes a deep learning-based computer-aided diagnosis (CAD) model, termed potent boosts channel-aware separable intent - ConvNeXt (PBCS-ConvNeXt), for automated NAFLD classification using B-mode US images. The model architecture comprises three key components: The potent stemcell, an advanced trainable preprocessing module for robust feature extraction; Enhanced ConvNeXt Blocks that amplify channel-wise features to refine processing; and the boosting block that integrates multi-stage features for effective information extraction from US data. Utilizing fatty liver gradings from attenuation imaging (ATI) as the ground truth, the PBCS-ConvNeXt model was evaluated using 5-fold cross-validation, achieving an accuracy of 82%, sensitivity of 81% and specificity of 83% for identifying fatty liver on abdominal US. The proposed CAD system demonstrates high diagnostic performance in NAFLD classification from US images, enabling early detection and informing timely clinical management to prevent disease progression.

Intelligent Intrusion Detection System Against Various Attacks Based on a Hybrid Deep Learning Algorithm.

The Internet of Things (IoT) has emerged as a crucial element in everyday life. The IoT environment is currently facing significant security concerns due to the numerous problems related to its architecture and supporting technology. In order to guarantee the complete security of the IoT, it is important to deal with these challenges. This study centers on employing deep learning methodologies to detect attacks. In general, this research aims to improve the performance of existing deep learning models. To mitigate data imbalances and enhance learning outcomes, the synthetic minority over-sampling technique (SMOTE) is employed. Our approach contributes to a multistage feature extraction process where autoencoders (AEs) are used initially to extract robust features from unstructured data on the model architecture's left side. Following this, long short-term memory (LSTM) networks on the right analyze these features to recognize temporal patterns indicative of abnormal behavior. The extracted and temporally refined features are inputted into convolutional neural networks (CNNs) for final classification. This structured arrangement harnesses the distinct capabilities of each model to process and classify IoT security data effectively. Our framework is specifically designed to address various attacks, including denial of service (DoS) and Mirai attacks, which are particularly harmful to IoT systems. Unlike conventional intrusion detection systems (IDSs) that may employ a singular model or simple feature extraction methods, our multistage approach provides more comprehensive analysis and utilization of data, enhancing detection capabilities and accuracy in identifying complex cyber threats in IoT environments. This research highlights the potential benefits that can be gained by applying deep learning methods to improve the effectiveness of IDSs in IoT security. The results obtained indicate a potential improvement for enhancing security measures and mitigating emerging threats.

A hybrid Framework for plant leaf disease detection and classification using convolutional neural networks and vision transformer

Recently, scientists have widely utilized Artificial Intelligence (AI) approaches in intelligent agriculture to increase the productivity of the agriculture sector and overcome a wide range of problems. Detection and classification of plant diseases is a challenging problem due to the vast numbers of plants worldwide and the numerous diseases that negatively affect the production of different crops. Early detection and accurate classification of plant diseases is the goal of any AI-based system. This paper proposes a hybrid framework to improve classification accuracy for plant leaf diseases significantly. This proposed model leverages the strength of Convolutional Neural Networks (CNNs) and Vision Transformers (ViT), where an ensemble model, which consists of the well-known CNN architectures VGG16, Inception-V3, and DenseNet20, is used to extract robust global features. Then, a ViT model is used to extract local features to detect plant diseases precisely. The performance proposed model is evaluated using two publicly available datasets (Apple and Corn). Each dataset consists of four classes. The proposed hybrid model successfully detects and classifies multi-class plant leaf diseases and outperforms similar recently published methods, where the proposed hybrid model achieved an accuracy rate of 99.24% and 98% for the apple and corn datasets.

Bridging the Gap: OCR Techniques for Noisy and Distorted Texts

Optical Character Recognition (OCR) has evolved significantly over the years, enabling automated text extraction from a variety of sources. However, OCR systems often struggle with noisy and distorted texts, such as those found in low-quality scans, degraded historical documents, or images captured in challenging conditions. This paper explores state-of-the-art techniques and advancements in OCR for handling noisy and distorted texts. We discuss preprocessing methods, robust feature extraction, deep learning models, and post-processing techniques, providing a comprehensive overview of the field. Additionally, we analyse gaps in current research and propose future directions for developing more resilient OCR systems.

Robust Finger Vein Presentation Attack Detection Using XceptionNet-based Modified Depthwise Separable Convolutional Neural Network

Finger vein presentation attack detection (FVPAD) biometric systems have seen substantial enhancements through the application of deep learning convolutional neural networks (DCNN). This advancement led to increased complexity, parameters and resource requirements. To address these challenges, a novel modification to the first entry flow of the XceptionNet architecture based on customized depthwise separable convolution (DSC) CNN-based for extracting robust features from FV images to detect spoofing attacks is proposed in this paper. The proposed approach stands out for its simplicity in design, fewer parameters, reduced computational load, minimal resource and equipment needs, and minimum data overflow while maintaining high accuracy in verification and classification tasks. The developed FVPAD system includes FV image data preprocessing and augmentation, a modified XceptionNet architecture based on DSC to deeply extract robust features. Finally, the fully connected (FC) layers exclusively use the SoftMax activation function to normalize, predict and classify output classes. The model was evaluated on cropped FV images from the IDIAP and SCUT-SFVD datasets, achieving high accuracy rates of 100% and 99.499%, respectively. It also has the lowest number of trainable parameters at 131,106 acquired from fifteen convolutional and depth-separable convolution layers.

Enhancing hyperspectral image object classification through robust feature extraction and spatial-spectral fusion using deep learning

Hyperspectral imaging (HSI) has gained significant attention in recent years due to its broad applications across agriculture, environmental monitoring, urban planning, infrastructure management, and defense and security for object detection and classification. Despite its potential, current methodologies face challenges such as insufficient feature extraction, noise interference, and inadequate spatial-spectral fusion, limiting classification accuracy and robustness. This study reviews advancements in HSI object detection and classification methodologies, emphasizing the role of machine-learning (ML) and deep-learning (DL) techniques. Hence, this work proposes a novel framework to address these challenges, prioritizing robust feature extraction, effective spatial-spectral fusion, and comprehensive noise removal mechanisms. By integrating DL techniques and training with HSI noisy data, this framework aims to enhance classification accuracy and robustness. The findings suggest that the proposed approach significantly improves the reliability and performance of HSI-based object classification systems. This research provides a pathway for future development in the domain, promising to elevate the effectiveness of HSI applications in real-world scenarios.

Leveraging Convolutional Neural Networks for Face Mask Detection

In response to the widespread use of face masks for safety, security, and health reasons, the development of automated systems for face mask detection has garnered significant attention. This research addresses binary face mask detection by leveraging Convolutional Neural Networks (CNNs) to create a specialized model. Using a Kaggle dataset, preprocessing steps including image resizing and color space conversion are applied for standardized data. The method entails a purpose-built CNN architecture comprising multiple convolutional layers with ReLU (Rectified Linear Unit) activations and max pooling for efficient feature extraction and spatial information capture. Further bolstering the architecture, fully connected layers coupled with dropout layers mitigate overfitting risks, enhancing generalization. The final sigmoid-activated output layer facilitates precise binary classification, distinguishing individuals with or without masks. Training guided by the Adam optimizer ensures parameter optimization based on accuracy metrics. This work contributes a meticulously designed CNN architecture optimized for face mask detection, showcasing robust feature extraction, spatial complexity handling, and overfitting mitigation, thereby presenting a potent solution with broad implications across industries for safety and public health measures. By aligning technology with societal needs, the research aids diverse industries in integrating automation and Artificial Intelligence for mask-wearing compliance. The findings underscore mask detection's pivotal role in overcoming challenges for safer environments.

Arrhythmia Classification Using CNN-SVM from ECG Spectrogram Representation

Arrhythmia, a critical subset of cardiovascular diseases and a leading cause of morbidity and mortality, is caused by irregular heartbeats that disrupt the normal rhythm of the heart. Detecting arrhythmias accurately is essential for timely diagnosis and treatment, which can be achieved through electrocardiogram (ECG) signals. This study presents a hybrid Convolutional Neural Network (CNN) and Support Vector Machine (SVM) model for arrhythmia classification, leveraging spectrogram representations of ECG signals. The CNN extracts spatial and temporal features from the spectrograms, while the SVM classifies five arrhythmia classes: Normal (N), Supra-ventricular premature (S), Ventricular escape (V), Fusion of ventricular and normal (F), and Unclassified (Q). Preprocessing techniques such as wavelet denoising and Short-Time Fourier Transform (STFT) were applied to improve signal quality and facilitate robust feature extraction. The proposed model was trained and evaluated on the MIT-BIH Arrhythmia Database, achieving a weighted F1-score of 0.985, demonstrating its ability to handle the imbalanced dataset effectively. Class-wise metrics highlighted high precision, recall, and F1-scores for majority classes and commendable performance for underrepresented classes, despite the inherent imbalance. These findings underscore the hybrid model's potential for arrhythmia classification by integrating the feature extraction strengths of CNNs with the precise classification capabilities of SVMs. Future research could address dataset imbalance through augmentation techniques and explore the model’s generalizability by testing on larger and more diverse datasets, paving the way for its application in real-world clinical scenarios.

MGFusion: a multimodal large language model-guided information perception for infrared and visible image fusion

Existing image fusion methods primarily focus on complex network structure designs while neglecting the limitations of simple fusion strategies in complex scenarios. To address this issue, this study proposes a new method for infrared and visible image fusion based on a multimodal large language model. The method proposed in this paper fully considers the high demand for semantic information in enhancing image quality as well as the fusion strategies in complex scenes. We supplement the features in the fusion network with information from the multimodal large language model and construct a new fusion strategy. To achieve this goal, we design CLIP-driven Information Injection (CII) approach and CLIP-guided Feature Fusion (CFF) strategy. CII utilizes CLIP to extract robust image features rich in semantic information, which serve to supplement the information of infrared and visible features, thereby enhancing their representation capabilities for the scene. CFF further utilizes the robust image features extracted by CLIP to select and fuse the infrared and visible features after the injection of semantic information, addressing the challenges of image fusion in complex scenes. Compared to existing methods, the main advantage of the proposed method lies in leveraging the powerful semantic understanding capabilities of the multimodal large language model to supplement information for infrared and visible features, thus avoiding the need for complex network structure designs. Experimental results on multiple public datasets validate the effectiveness and superiority of the proposed method.

Performance Evaluation of Various Deep Learning Models in Gait Recognition Using the CASIA-B Dataset

Human gait recognition (HGR) has been employed as a biometric technique for security purposes over the last decade. Various factors, including clothing, carrying items, and walking surfaces, can influence the performance of gait recognition. Additionally, identifying individuals from different viewpoints presents a significant challenge in HGR. Numerous conventional and deep learning techniques have been introduced in the literature for HGR, but traditional methods are not well suited to handling large datasets. This research explores the effectiveness of four deep learning models for gait identification in the CASIA B dataset: the convolutional neural network (CNN), multi-layer perceptron (MLP), self-organizing map (SOMs), and transfer learning with EfficientNet. The selected deep learning techniques offer robust feature extraction and the efficient handling of large datasets, making them ideal in enhancing the accuracy of gait recognition. The collection includes gait sequences from 10 individuals, with a total of 92,596 images that have been reduced to 64 × 64 pixels for uniformity. A modified model was developed by integrating sequential convolutional layers for detailed spatial feature extraction, followed by dense layers for classification, optimized through rigorous hyperparameter tuning and regularization techniques, resulting in an accuracy of 97.12% for the test set. This work enhances our understanding of deep learning methods in gait analysis, offering significant insights for choosing optimal models in security and surveillance applications.

Bridging the Gap Between Computational Efficiency and Segmentation Fidelity in Object-Based Image Analysis.

A critical issue in image analysis for analyzing animal behavior is accurate object detection and tracking in dynamic and complex environments. This study introduces a novel preprocessing algorithm to bridge the gap between computational efficiency and segmentation fidelity in object-based image analysis for machine learning applications. The algorithm integrates convolutional operations, quantization strategies, and polynomial transformations to optimize image segmentation in complex visual environments, addressing the limitations of traditional pixel-level and unsupervised methods. This innovative approach enhances object delineation and generates structured metadata, facilitating robust feature extraction and consistent object representation across varied conditions. As empirical validation shows, the proposed preprocessing pipeline reduces computational demands while improving segmentation accuracy, particularly in intricate backgrounds. Key features include adaptive object segmentation, efficient metadata creation, and scalability for real-time applications. The methodology's application in domains such as Precision Livestock Farming and autonomous systems highlights its potential for high-accuracy visual data processing. Future work will explore dynamic parameter optimization and algorithm adaptability across diverse datasets to further refine its capabilities. This study presents a scalable and efficient framework designed to advance machine learning applications in complex image analysis tasks by incorporating methodologies for image quantization and automated segmentation.

Face Forgery Detection Using Deep Learning

The rising occurrence of forged content poses a threat to the authenticity of multimedia. This research suggests a simplified hybrid architecture as an effective method for detecting face forgeries. The framework includes CNN for dependable classification, EfficientNet for robust feature extraction, and MTCNN for precise face detection. MTCNN ensures that high-quality input is produced for feature extraction by accurately localizing facial regions. The CNN classifier utilizes the extracted features in order to distinguish between authentic and manipulated content, and EfficientNet, which has become famous for its good performance and computational efficiency, is able to capture face patterns at a subtle level. Transfer learning enhances the adaptability of the model toward new manipulation techniques because it pre- trains on a large-scale dataset before fine-tuning on data specifically related to deepfakes. Key Words: Deep Learning, Deepfakes, Face Forgery, Multimedia forensics, CNN.

Research Advanced in Image Inpainting based on Deep Learning

Image inpainting has always been a hotspot in the computer vision community, which aims to restore damaged or missing parts of an image, ensuring that the restored image is physically reasonable, visually pleasing, and consistent in texture and structure with the original image. Early inpainting methods primarily depend on diverse image processing technologies, such as sample replication, partial differential equation solving, etc. As artificial intelligence rapidly progresses, image inpainting methods based on deep learning exploit robust feature extraction capabilities to extract useful information from vast amounts of data, contributing to more robust and natural inpainting effects, especially when dealing with complex and high-dimensional scenes. This paper provides a review of deep learning-based image inpainting technologies, categorized into four main types: autoencoders, generative adversarial networks (GANs), recurrent neural networks (RNNs), and convolutional neural networks (CNNs). It summarizes and analyzes the basic principles, advantages, disadvantages, and performance of these methods in image inpainting, as well as introduces commonly used datasets and quantitative evaluation metrics. Additionally, the paper discusses the principal challenges currently faced in this domain and provides insights into future research directions for deep learning-based image inpainting.

Hybrid 3D Convolutional–Transformer Model for Detecting Stereotypical Motor Movements in Autistic Children During Pre-Meltdown Crisis

Computer vision using deep learning algorithms has served numerous human activity identification applications, particularly those linked to safety and security. However, even though autistic children are frequently exposed to danger as a result of their activities, many computer vision experts have shown little interest in their safety. Several autistic children show severe challenging behaviors such as the Meltdown Crisis which is characterized by hostile behaviors and loss of control. This study aims to introduce a monitoring system capable of predicting the Meltdown Crisis condition early and alerting the children’s parents or caregivers before entering more difficult settings. For this endeavor, the suggested system was constructed using a combination of a pre-trained Vision Transformer (ViT) model (Swin-3D-b) and a Residual Network (ResNet) architecture to extract robust features from video sequences to extract and learn the spatial and temporal features of the Stereotyped Motor Movements (SMMs) made by autistic children at the beginning of the Meltdown Crisis state, which is referred to as the Pre-Meltdown Crisis state. The evaluation was conducted using the MeltdownCrisis dataset, which contains realistic scenarios of autistic children’s behaviors in the Pre-Meltdown Crisis state, with data from the Normal state serving as the negative class. Our proposed model achieved great classification accuracy, at 92%.

An AI-Driven Framework for Detecting Bone Fractures in Orthopedic Therapy.

This study presents an advanced artificial intelligence-driven framework designed to enhance the speed and accuracy of bone fracture detection, addressing key limitations in traditional diagnostic approaches that rely on manual image analysis. The proposed framework integrates the YOLOv8 object detection model with a ResNet backbone to combine robust feature extraction and precise fracture classification. This combination effectively identifies and categorizes bone fractures within X-ray images, supporting reliable diagnostic outcomes. Evaluated on an extensive data set, the model demonstrated a mean average precision of 0.9 and overall classification accuracy of 90.5%, indicating substantial improvements over conventional methods. These results underscore a potential framework to provide healthcare professionals with a powerful, automated tool for orthopedic diagnostics, enhancing diagnostic efficiency and accuracy in routine and emergency care settings. The study contributes to the field by offering an effective solution for automated fracture detection that aims to improve patient outcomes through timely and accurate intervention.

DyPanVO: a robust monocular visual odometry in dynamic environments by utilizing multiple deep neural networks

Abstract Traditional monocular Visual Odometry (VO) is typically based on the assumption of a static environment. However, it performs poorly in dynamic scenes, suffering from error accumulation and scale drift. To address these issues, a novel framework, named DyPanVO, was proposed, which incorporates multiple deep neural networks for feature point extraction, panoptic segmentation, and depth estimation. Firstly, learning-based methods are employed, specifically SuperPoint for robust feature extraction and LightGlue for accurate feature matching. Then, outlier points are eliminated by combining dynamic prior results from panoptic segmentation with epipolar geometry constraints to determine the motion state of objects. Additionally, the introduction of depth information ensures that scale is recovered and fundamentally solves the issue of error accumulation. Two types of experiment (outdoor and indoor scenes) are carried out to show the effectiveness of DyPanVO. Moreover, the comparison results demonstrate that it performs comparably with multi-view geometry-based and outperforms learning-based methods.

Outlier Handling Strategy of Ensembled-Based Sequential Convolutional Neural Networks for Sleep Stage Classification.

The pivotal role of sleep has led to extensive research endeavors aimed at automatic sleep stage classification. However, existing methods perform poorly when classifying small groups or individuals, and these results are often considered outliers in terms of overall performance. These outliers may introduce bias during model training, adversely affecting feature selection and diminishing model performance. To address the above issues, this paper proposes an ensemble-based sequential convolutional neural network (E-SCNN) that incorporates a clustering module and neural networks. E-SCNN effectively ensembles machine learning and deep learning techniques to minimize outliers, thereby enhancing model robustness at the individual level. Specifically, the clustering module categorizes individuals based on similarities in feature distribution and assigns personalized weights accordingly. Subsequently, by combining these tailored weights with the robust feature extraction capabilities of convolutional neural networks, the model generates more accurate sleep stage classifications. The proposed model was verified on two public datasets, and experimental results demonstrate that the proposed method obtains overall accuracies of 84.8% on the Sleep-EDF Expanded dataset and 85.5% on the MASS dataset. E-SCNN can alleviate the outlier problem, which is important for improving sleep quality monitoring for individuals.

ECMHA-PP: A Breast Cancer Prognosis Prediction Model Based on Energy-Constrained Multi-Head Self-Attention.

Breast cancer is a significant threat to women's health. Precise prognosis prediction for breast cancer can help doctors implement more rational treatment strategies. Artificial intelligence can assist doctors in decision-making and enhance predictionaccuracy. In this paper, a deep learning model ECMHA-PP (Energy Constrained Multi-Head Self-Attention based Prognosis Prediction) is proposed to predict the prognosis of breast cancer. ECMHA-PP utilizes patients' clinical data and extracts features through a cross-position mix and a channel mix multi-layer perceptron. Then, it incorporates an energy-constrained multi-head self-attention layer to improve feature extraction capability. The source code of ECMHA-PP has been hosted on GitHub and is available at https://github.com/xiaoliu166370/ECMHA-PP. To evaluate our proposed method, prognostic prediction experiments were performed on the METABRIC dataset, yielding outstanding results with an average accuracy of 93.0% and an average area under the curve of 0.974. To further validate the model's performance, we conducted tests on another independent dataset, BRCA, achieving an accuracy of 87.6%. In comparison with other widely used advanced methods, ECMHA-PP demonstrated higher comprehensive performance, making it a reliable prognostic prediction model for breast cancer. Given its robust feature extraction and predictioncapabilities.

Fast Implementation of Image Preprocessing for SAR Scene Matching Guidance

Abstract The design of SAR scene matching guidance scheme and the hardware allocation of algorithm module are discussed in this paper. Aiming at the requirement of SAR image preprocessing module, a fast robust feature extraction preprocessing method based on FPGA is designed, and the fast implementation of Gabor filter and wavelet transform is completed. Through the hardware-in-the-loop simulation module test and experimental verification, the FPGA-based pre-processing fast implementation is about 60 times shorter than the DSP processing time, and the results of FPGA, DSP and digital simulation in the hardware-in-the-loop simulation process are consistent. The proposed method not only shortens the processing time of scene matching guidance system, but also ensures the accuracy of SAR image matching.

A Multi-scale Patch Mixer Network for Time Series Anomaly Detection

With the development of Internet of Things (IoT) technology, a large amount of data with temporal characteristics is collected and stored. How to efficiently and accurately identify anomalies from these data is a major challenge. At present, there are many problems in the application of anomaly detection, including non-stationary data, complex and difficult-to-collect anomalies, the need for real-time detection and the limitation of computing resources. But few methods can comprehensively consider these issues. To overcome these challenges, we propose a lightweight neural network, Multi-scale Patch Mixer Network (MP-MixerNet). It is mainly composed of a Mixer Block based on fully connected layer design, which contains a Temporal-Mixer and a Spatial-Mixer, and can simultaneously model the intra- and inter-series dependencies of multivariate time series. We also perform multi-scale patch segmentation based on frequency analysis, which helps the model extract robust features from multiple period views. In addition, we design an Input Stabilization module to help the model deal with data distribution shift. Experimental results on a public time series anomaly detection dataset show that we are able to achieve higher comprehensive performance with fewer parameters and inference time.