Convolutional Neural Network Classifier Research Articles

A light-weight factorized convolutions based dual-input fuzzy-CNN for efficient motor bearing fault diagnosis

ABSTRACT Efficient and timely identification of bearing faults is imperative to ensure operational normalcy, reduced down-times and health hazards in motor fault tolerant control systems. This paper proposes a fault diagnosis method that combines the vibration information with time-varying rotational speed for effective fault diagnosis under non-stationary conditions. The complex wavelet transform is used to encode both the vibrational and rotational signals in 2d representations for spatial feature extraction. An efficient algorithm is proposed to select the mother wavelet with the least average entropy. Moreover, a spatial decomposition-based approach using factorised convolutions is used to create a light-weight fuzzy convolutional neural network named Split-Operation Fuzzy Convolutional Neural Network (SOF-CNN) for semi rule-based feature extraction and classification. The performance was evaluated on the University of Ottawa (UOO) dataset for multiple speed conditions and cross-condition validation with the highest accuracy yield being 99.92% from the fourth condition and 3rd trial acquired via averaged 10-Fold Cross Validation. The average accuracy yield across all the scenario was 99.89% with 64.73% being the highest accuracy for cross-condition validation. The performance was evaluated across a diverse range of evaluation criterion including both quantitative and statistical tests.

Comparative Analysis of Conventional CNN v’s ImageNet Pretrained ResNet in Medical Image Classification

Convolutional Neural Networks (CNNs) are the prevalent technology in computer vision and have become increasingly popular for medical imaging data classification and analysis. In this field, due to the scarcity of medical data, pretrained ResNets on ImageNet can be considered a suitable first approach. This paper examines the medical imaging classification accuracy of conventional basic custom CNNs compared to ImageNet pretrained ResNets on various medical datasets in an effort to give more information about the importance of medical data and its preprocessing techniques for disease studies. Microscope-extracted cytological images were examined along with chest X-rays, MRI brain scans, and melanoma photographs. The medical images were examined in various sets, class combinations, and resolutions. Augmented image datasets and asymmetrical training and validation splits among the classes were also examined. Models were developed after they were tested and fine-tuned with respect to their network size, parameter values and network methods, image resolution, size of dataset, multitude, and genre of class. Overfitting was also examined, and comparative studies regarding the computational cost of different models were performed. The models achieved high accuracy in image classification that varies depending on the dataset and can be easily incorporated in future over-the-internet medical decision-supporting (telemedicine) environments. In addition, it appeared that conventional basic custom CNN overperformed ImageNet pretrained ResNets. The obtained results indicate the importance of utilizing medical image data as a testbed for improvements in CNN classification performance and the possibility of using CNNs and data preprocessing techniques for disease studies.

Voice Analysis in Dogs with Deep Learning: Development of a Fully Automatic Voice Analysis System for Bioacoustics Studies

Extracting behavioral information from animal sounds has long been a focus of research in bioacoustics, as sound-derived data are crucial for understanding animal behavior and environmental interactions. Traditional methods, which involve manual review of extensive recordings, pose significant challenges. This study proposes an automated system for detecting and classifying animal vocalizations, enhancing efficiency in behavior analysis. The system uses a preprocessing step to segment relevant sound regions from audio recordings, followed by feature extraction using Short-Time Fourier Transform (STFT), Mel-frequency cepstral coefficients (MFCCs), and linear-frequency cepstral coefficients (LFCCs). These features are input into convolutional neural network (CNN) classifiers to evaluate performance. Experimental results demonstrate the effectiveness of different CNN models and feature extraction methods, with AlexNet, DenseNet, EfficientNet, ResNet50, and ResNet152 being evaluated. The system achieves high accuracy in classifying vocal behaviors, such as barking and howling in dogs, providing a robust tool for behavioral analysis. The study highlights the importance of automated systems in bioacoustics research and suggests future improvements using deep learning-based methods for enhanced classification performance.

Implementasi Deep Learning Untuk Menentukan Harga Buah Sawit

This study aims to analyze the price of palm oil using Convolutional Neural Network (CNN) method in deep learning. CNN was chosen for its ability to process complex data and recognize patterns from diverse data. The stages of research include data analysis, data pre-processing, predictive model design for CNN method, CNN classification model prediction results, CNN method evaluation, and CNN method evaluation results. This study aims to produce a model that can predict the price of oil palm with high accuracy, based on data covering a variety of characteristics of farmers and the quality of oil palm crops. Prediction results were conducted using data from 50 oil palm farmers. From the prediction, as many as 23 data farmers get a price of IDR 2,300, 13 other farmers get a price of IDR 2,000, and the remaining 14 data farmers get a price of IDR 1,800. The results of this prediction are based on data from farmers and the quality of oil palm crops they grow and produce. By utilizing the CNN method, the model can capture various factors that affect the price of palm oil, including the quality of palm fruit and agricultural conditions. Evaluation of the CNN method showed very good results, with almost perfect accuracy. This method managed to predict palm oil prices very precisely, showing that CNN can be an effective tool in the analysis of palm oil prices. The results of this evaluation confirmed that the CNN method can be relied upon to provide accurate predictions, helping farmers and palm oil industry players in determining prices that are in accordance with the quality and condition of the crop.

HFF-Net: A hybrid convolutional neural network for diabetic retinopathy screening and grading

Diabetic Retinopathy (DR) is a leading cause of vision loss among diabetic patients, necessitating effective screening and grading for timely intervention. Regular screening significantly increases the workload of ophthalmologists, and accurate grading into stages—mild, moderate, severe, and proliferative—is crucial for monitoring disease progression. While Computer-Aided Diagnosis (CAD) systems can alleviate this burden, existing Convolutional Neural Network (CNN)-based frameworks use fixed-size kernels in a linear feed-forward manner. This approach can lead to information loss in the initial stages due to limited feature utilization across adjacent layers. To address this limitation, we propose a Hierarchical Features Fusion Convolutional Neural Network (HFF-Net) within a Diabetic Retinopathy Screening and Grading (DRSG) framework. The framework includes preprocessing to extract regions of interest from fundus images (FIs), enhancement using Contrast Limited Adaptive Histogram Equalization (CLAHE), and data augmentation for class balancing and overfitting mitigation. HFF-Net extracts multiscale features that fused at multiple levels within the network, utilizing the swish activation function for improved learning stability. We evaluated HFF-Net against several state-of-the-art CNN classifiers within the DRSG framework. Experimental results demonstrate that HFF-Net achieves a grading accuracy of 73.77 ​%, surpassing the second-best model by 3.51 percentage points (a relative improvement of approximately 5 ​%), and attains a screening accuracy of 98.70 ​% using only 1.18 million parameters. These findings highlight HFF-Net's potential as an efficient and effective tool in CAD systems for DR screening and grading.©2017 Elsevier Inc. All rights reserved.

Enhanced driving assistance: automated day and night vehicle detection system utilizing convolutional neural networks

<p>This paper presents an enhanced real-time vehicle detection system using convolutional neural networks (CNNs) for both daytime and night-time conditions. Initially, the system determines the time of capture by analyzing the upper part of input images. For daytime detection, it uses normalized cross-correlation and two-dimensional discrete wavelet transform (2D-DWT) techniques. Night-time detection involves identifying vehicle lamps through color thresholding and connected component techniques, followed by symmetry analysis and CNN classification. The dataset for training includes images from the Caltech Cars, AOLP, KITTI Vision, and night-time vehicle detection datasets, ensuring robust performance across various lighting conditions. Experiments demonstrate the system's high accuracy, achieving 99.2% during the day and 98.27% at night, meeting real-time requirements and enhancing driving assistance systems' reliability.</p>

Detecting student attention through electroencephalography signals: a comparative analysis of deep learning models

<span lang="EN-MY">In the landscape of educational technology, understanding and optimizing student attention is important to enhance student’s learning experience. This study explores the potential of using electroencephalography (EEG) signals for discerning students' attention levels during educational tasks. With a cohort of 30 participants, EEG data were meticulously collected and subjected to robust preprocessing techniques, including independent component analysis (ICA) and principal component analysis (PCA). The research then employed different deep learning algorithm such as long short-term memory (LSTM), recurrent neural network (RNN), gated recurrent unit (GRU), multi-layer perceptron (MLP), and convolutional neural network (CNN) classifiers to predict students' attention. The results reveal notable variations in the classifiers' predictive performance. Our finding revealed that the LSTM model emerged as the top performer and achieved 96% of the accuracy. This study not only contributes to the advancement of attention detection in educational technology but also underscores the importance of preprocessing methodologies, such as ICA and PCA, in optimizing the performance of deep learning models for EEG-based applications.</span>

The Implementation of Bayesian Optimization for Automatic Parameter Selection in Convolutional Neural Network for Lung Nodule Classification

Lung cancer's high mortality rate makes early detection crucial. Machine learning techniques, especially convolutional neural networks (CNN), play a very important role in lung nodule detection. Traditional machine learning approaches often require manual feature extraction, while CNNs, as a specialized type of neural network, automatically learn features directly from the data. However, tuning CNN hyperparameters, such as network structure and training parameters, is computationally intensive. Bayesian Optimization offers a solution by efficiently selecting parameter values. This study develops a CNN classification model with hyperparameter tuning using Bayesian Optimization, achieving a 97.2% accuracy. Comparatively, Particle Swarm Optimization and Genetic Algorithm methods each resulted in 96.4% accuracy. The research concludes that Bayesian Optimization is an effective approach for CNN hyperparameter tuning in lung nodule classification.

Seeding precision: a mask region based convolutional neural networks classification approach for the classification of paddy seeds

The generation of sufficient training data that is accurately labelled for a deep neural network involves a significant amount of effort and frequently constitutes a bottleneck in the implementation process. For the purpose of this research, we are training a neural network model to perform instance segmentation and classification of crop seeds for various rice cultivars. Synthetically constructed dataset is used here. The concept of domain randomization, which offers a productive alternative to the laborious process of data annotation, serves as the basis for our methodology. We make use of the domain randomization technique in order to produce synthetic data, and the mask region-based convolutional neural network (Mask R-CNN) architecture is utilized in order to train our neural network models. A cultivar name is used to designate the seeds, and they are differentiated from one another using colors that are comparable to those used in the actual dataset of paddy cultivars. Our mission focuses on the identification and categorization of rice paddy varieties within automatically generated photographs. Farmers are able to accurately sort crop seeds from a variety of rice cultivars with the use of this approach, which is particularly useful for phenotyping and optimizing yields in laboratory settings.

Improving baseline reduction for emotion recognition based on electroencephalogram signals

The baseline reduction method has been widely used to define electroencephalogram (EEG) signal patterns. However, because the baseline signal in this approach contains artifacts, the baseline reduction approach cannot perform optimally. As a result, decreasing artifacts in the baseline signal is critical. The mean, Gaussian, and Savitzky-Golay filters will be compared in this study to minimize artifacts in the baseline signal. Three secondary datasets are utilized to evaluate these approaches' capacity to remove artifacts. These three strategies are also tested with the convolution neural network classification algorithms. When applied to the dataset for emotion analysis using physiological signals (DEAP) and a dataset for multimodal research of affect, personality traits, and mood on individuals and groups (AMIGOS) datasets, the mean filter can increase baseline reduction performance based on twenty-four test scenarios. On the data readiness for machine learning research (DREAMER) dataset, however, the Gaussian filter is preferable. The relative difference approach was employed in this study's baseline reduction process to generate EEG signal patterns that are easy to recognize throughout the classification phase, which impacts increasing accuracy.

Water Trash Detection System: An Innovative Approach to Monitoring Aquatic Pollution

Abstract: Water pollution, particularly from non-biodegradable waste such as plastics, is one of the most pressing environmental issues facing aquatic ecosystems today. This paper presents an innovative solution for detecting and classifying waste in water bodies using machine learning techniques, specifically You Only Look Once (YOLO) and Convolutional Neural Networks (CNNs). The Water Trash Detection System (WTDS) can process both static images and real-time video streams to efficiently detect various types of aquatic trash, including plastic bottles, bags, and fishing debris. The system generates detailed reports on detected waste, providing environmental organizations and researchers with actionable data for monitoring and cleaning polluted water bodies. The combination of YOLO for real-time object detection and CNNs for waste classification allows the WTDS to deliver high accuracy and real-time processing, which is essential for effective environmental management.

Automatic Detection and Classification of Natural Weld Defects Using Alternating Magneto-Optical Imaging and ResNet50.

It is difficult to detect and identify natural defects in welded components. To solve this problem, according to the Faraday magneto-optical (MO) effect, a nondestructive testing system for MO imaging, excited by an alternating magnetic field, is established. For the acquired MO images of crack, pit, lack of penetration, gas pore, and no defect, Gaussian filtering, bilateral filtering, and median filtering are applied for image preprocessing. The effectiveness of these filtering methods is evaluated using metrics such as peak signal-noise ratio (PSNR) and mean squared error. Principal component analysis (PCA) is employed to extract column vector features from the downsampled defect MO images, which then serve as the input layer for the error backpropagation (BP) neural network model and the support vector machine (SVM) model. These two models can be used for the classification of partial defect MO images, but the recognition accuracy for cracks and gas pores is comparatively low. To further enhance the classification accuracy of natural weld defects, a convolutional neural network (CNN) classification model and a ResNet50 classification model for MO images of natural weld defects are established, and the model parameters are evaluated and optimized. The experimental results show that the overall classification accuracy of the ResNet50 model is 99%. Compared with the PCA-SVM model and CNN model, the overall classification accuracy was increased by 7.4% and 1.8%, and the classification accuracy of gas pore increased by 10% and 4%, respectively, indicating that the ResNet50 model can effectively and accurately classify natural weld defects.

Adaptive wireless-powered network based on CNN near-field positioning by a dual-band metasurface

With the improvement of industry, the connectivity of electronic devices gradually shift from wired to wireless. As a solution for power delivery, the non-contact power transfer holds promising ways to charge for moving terminals, enabling battery-free sensing, processing, and communication. Based on a dual-band metasurface, this study proposes an adaptive wireless-powered network (AWPN) to realize the simultaneous wireless localization and non-contact power supply. It first achieves localization with 3 cm resolution on a single-input single-output (SISO) system, by combining space-time-coding (STC) and convolutional neural network (CNN). With precise position information, AWPN real-time aligns power beams to the terminals for stable energy transmission. Then, battery-free terminals enable to perceive the environmental data and uploads the results. From the measurement results, AWPN gets more than 98% CNN classification accuracy and can tolerate certain environmental changes. Thus, being adaptive and contactless, our study will propel the advancement in Internet of Things (IoT), intelligent metasurface, and the robot industry.

A Review of Convolutional Neural Networks in Cancer Image Classification

Abstract. Cancer image analysis is a crucial element in the field of modern oncology, offering insights into tumor characteristics that are vital for accurate diagnosis and treatment planning. Conventional diagnostic techniques are frequently constrained by discrepancies in interpretation and delays in processing. In recent years, deep learning techniques, in particular convolutional neural networks (CNNs), have demonstrated remarkable capabilities in processing large-scale medical images, enabling the rapid detection and classification of cancerous tissue with high accuracy and precision. This paper reviews the application of CNNs in cancer image classification, introduces common CNN models, and discusses advances such as transfer learning and multimodal data fusion. Furthermore, this paper examines the current challenges confronting convolutional neural networks (CNNs) in the domain of cancer classification. It proposes promising avenues for future research and underscores the potential of CNNs in enhancing cancer diagnosis and even revolutionizing the healthcare system.

Quantum deep learning in Parkinson’s disease prediction using hybrid quantum–classical convolution neural network

Quantum deep learning in Parkinson’s disease prediction using hybrid quantum–classical convolution neural network

Atrial Fibrillation Type Classification by a Convolutional Neural Network Using Contrast-Enhanced Computed Tomography Images

Catheter ablation therapy, which is a treatment for atrial fibrillation (AF), has a higher recurrence rate as AF duration increases. Compared to paroxysmal AF (PAF), sustained AF is known to cause progressive anatomic remodeling of the left atrium, resulting in enlargement and shape changes. In this study, we used contrast-enhanced computed tomography (CT) to classify atrial fibrillation (AF) into paroxysmal atrial fibrillation (PAF) and long-term persistent atrial fibrillation (LSAF), which have particularly different recurrence rates after catheter ablation. Contrast-enhanced CT images of 30 patients with PAF and 30 patients with LSAF were input into six pretrained convolutional neural networks (CNNs) for the binary classification of PAF and LSAF. In this study, we propose a method that can recognize information regarding the body axis direction of the left atrium by inputting five slices near the left atrium. The classification was visualized by obtaining a saliency map based on score-class activation mapping (CAM). Furthermore, we surveyed cardiologists regarding the classification of AF types, and the results of the CNN classification were compared with the results of physicians’ clinical judgment. The proposed method achieved the highest correct classification rate (81.7%). In particular, models with shallow layers, such as VGGNet and ResNet, are able to capture the overall characteristics of the image and therefore are likely to be suitable for focusing on the left atrium. In many cases, patients with an enlarged left atrium tended to have long-lasting AF, confirming the validity of the proposed method. The results of the saliency map and survey of physicians’ basis for judgment showed that many patients tended to focus on the shape of the left atrium in both classifications, suggesting that this method can classify atrial fibrillation more accurately than physicians, similar to the judgment criteria of physicians.

Separable Integral Neural Networks

Integral neural networks adopt continuous integral operators instead of conventional discrete convolutional operations to perform deep learning tasks. As this integral operator is the continuous representation of the regular convolutional operation, it is not suitable for representing the separable convolutional operations widely deployed on mobile devices. To address this issue, a separable integral layer composed of a depth-wise integral operator and a point-wise integral operator is proposed in this paper to represent discrete depth-wise and point-wise convolutional operations in continuous manner. According to the fabric units of five classical convolutional neural networks(NIN, VGG11, GoogleNet, ResNet18, ResNet50), we design five kinds of separable integral blocks(SIBs) to encapsulate separable integral layers in different manner. Using the proposed SIBs as basic blocks, a family of lightweight separable integral neural networks(SINNs) are constructed and deployed on resource-constrained mobile devices. SINNs have the characteristics of integral neural networks, i.e., performing structural pruning without fine-tuning, and also inherit the advantages of separable convolutional operations, i.e., reducing the computational cost while keeping a competitive performance. The experimental results show that SINNs achieve the similar performance with the state-of-the-art integral neural networks(INNs), while reducing the computational cost to up to 1/1.79 times that of INN(1.74× fewer parameters than INN using ResNet101 backbone framework) on ImageNet dataset. The code will be released at https://github.com/ljh3832-ccut/SINN.

Convolutional neural network classification of ultrasound parametric images based on echo-envelope statistics for the quantitative diagnosis of liver steatosis.

Early detection and quantitative evaluation of liver steatosis are crucial. Therefore, this study investigated a method for classifying ultrasound images to fatty liver grades based on echo-envelope statistics (ES) and convolutional neural network (CNN) analyses. Three fatty liver grades, i.e., normal, mild, and moderate-to-severe, were defined using the thresholds of the magnetic resonance imaging-derived proton density fat fraction (MRI-PDFF). There were 10 cases of each grade, totaling 30 cases. To visualize the texture information affected by the deposition of fat droplets within the liver, the maps of first- and fourth-order moments and the heat maps formed from both moments were employed as parametric images derived from the ES. Several dozen to hundreds of regions of interest (ROIs) were extracted from the liver region in each parametric image. A total of 7680 ROIs were utilized for the transfer learning of a pretrained VGG-16 and classified using the transfer-learned VGG-16. The classification accuracies of the ROIs in all types of the parametric images were approximately 46%. The fatty liver grade for each case was determined by hard voting on the classified ROIs within the case. In the case of the fourth-order moment maps, the classification accuracy of the cases through hard voting mostly increased to approximately 63%. The formation of parametric images derived from the ES and the CNN classification of the parametric images were proposed for the quantitative diagnosis of liver steatosis. In more than 60% of the cases, the fatty liver grade could be estimated solely using ultrasound images.

Optimization of energy acquisition system in smart grid based on artificial intelligence and digital twin technology

In response to the low operating speed and poor stability of energy harvesting systems in smart grids, an energy harvesting optimization method based on improved convolutional neural networks and digital twin technology is proposed in the experiment. Firstly, a smart grid data transmission framework integrating digital twin technology is proposed. A digital twin mapping method based on time, data, and topology structure is used to realize the digital twin mapping at the device level of power grid. Through data synchronization and interaction between the physical power grid and the digital twin model, the operational efficiency and reliability of the power grid are improved. Then, the classical convolutional neural network and attention mechanism are used to comprehensively analyze the physical topology data in the smart grid energy acquisition system. The improved lightweight target detection model is combined to monitor the equipment status of the smart grid and extract key features. Simultaneously utilizing convolutional attention mechanism to dynamically adjust the feature weights of channels or spaces, completing the preprocessing of energy harvesting data. Finally, combined with energy harvesting and power grid switching system, the process of energy harvesting and power grid operation are optimized together. On the training and validation sets, when the channels exceeded 60, the proposed method achieved a system energy efficiency of 55% during operation. The system energy efficiency of the other three comparative algorithms was all less than 40%. In practical applications, as the energy transfer loss increased to 1.0, the system throughput increased to 50 bits. The electricity needs of different users were met, and the difference between power allocation and optimal power allocation was small, which was very reasonable. This proves that the research has effectively optimized the energy harvesting system in the smart grid, improving the efficiency and reliability of the system in practical applications of the smart grid. At the same time, in the increasingly severe energy problem, this system can further provide technical references for the utilization of renewable energy and help achieve the goal of sustainable energy.

Combining Endpoint Detection with a Convolutional Neural Network Classifier for the Automatic Recognition of Cardiac Arrhythmias in Electrocardiogram Signals

Combining Endpoint Detection with a Convolutional Neural Network Classifier for the Automatic Recognition of Cardiac Arrhythmias in Electrocardiogram Signals