Feature Classification Research Articles

Research on Pedestrian and Cyclist Classification Method Based on Micro-Doppler Effect.

In the field of autonomous driving, it is important to protect vulnerable road users (VRUs) and ensure the safety of autonomous driving effectively by improving the detection accuracy of VRUs in the driver's field of vision. However, due to the strong temporal similarity between pedestrians and cyclists, the insensitivity of the traditional least squares method to their differences results in its suboptimal classification performance. In response to this issue, this paper proposes an algorithm for classifying pedestrian and cyclist targets based on the micro-Doppler effect. Firstly, distinct from conventional time-frequency fusion methods, a preprocessing module was developed to solely perform frequency-domain fitting on radar echo data of pedestrians and cyclists in forward motion, with the purpose of generating fitting coefficients for the classification task. Herein, wavelet threshold processing, short-time Fourier transform, and periodogram methods are employed to process radar echo data. Then, for the heightened sensitivity to inter-class differences, a fractional polynomial is introduced into the extraction of micro-Doppler characteristics of VRU targets to enhance extraction precision. Subsequently, the support vector machine technique is embedded for precise feature classification. Finally, subjective comparisons, objective explanations, and ablation experiments demonstrate the superior performance of our algorithm in the field of VRU target classification.

A defect classification algorithm for gas tungsten arc welding process based on unsupervised learning and few-shot learning strategy

Welding defect prediction is the foundation for ensuring welding quality in gas tungsten arc welding (GTAW). In the prediction process, method based on molten pool vision is the most effective. Since the classification of molten pool defects relies on a substantial volume of labeled data, it is challenging for the models to be applied industrially. This paper presents an algorithm, FS-Classifier, that can achieve high prediction accuracy based on a limited amount of labeled data. The FS-Classifier comprises two stages: Firstly, an unsupervised training approach named RaP is designed to pre-train the feature extractor using extensive unlabeled daily datasets. The RaP consists of a rotation angle prediction task and a position prediction task, which ensure that the network focuses on salient features and precise elements, respectively. Secondly, the support vectors constructed from limited labeled data are used for the feature classifier. The input data is classified to certain class by computing its distances to support vector. The model achieves an accuracy of 94.5 % on the private dataset and 92.8 % on the public dataset for the six classes of defects using 5 % of labeled data volume. In addition, comparative experiments show that our method only requires 5 % of labeled data to achieve accuracy comparable to traditional supervised learning methods. The proposed algorithm addresses the issue of relying on a substantial amount of labeled data in welding process defect classification.

Extraction and classification of vibration features of rolling element bearing with increasing the rotational speed

Machinery components degrade over time due to continuous use. A reliable prognosis framework can improve machinery health by monitoring the behavior of its parts and providing warnings before critical failures occur. Bearings, which are essential components of rotating machinery, help maintenance personnel assess the machine’s condition during continuous wear. In this study, vibration data from roller bearings under various conditions and faults were collected. The Vibration analysis technique was employed to detect and classify different faults in bearings based on the characteristics of the vibration signals generated by the machinery. Faults can be detected, diagnosed, and classified by analyzing bearing vibration signatures using techniques such as frequency analysis, time-domain analysis, spectral analysis, and kurtogram classifiers. This enables appropriate maintenance actions to be taken in time, preventing further damage or failures.

Few-shot learning based histopathological image classification of colorectal cancer

Background Colorectal cancer is a prevalent and deadly disease worldwide, posing significant diagnostic challenges. Traditional histopathologic images classification is often inefficient and subjective. Although some histopathologists use computer-aided diagnosis (CAD) to improve efficiency, these methods depend heavily on extensive data and specific annotations, limiting their development. To address these challenges, this paper proposes a method based on few-shot learning. Methods This study introduces a few-shot learning approach that combines transfer learning and contrastive learning to classify colorectal cancer histopathology images into benign and malignant categories. The model comprises modules for feature extraction, dimensionality reduction, and classification, trained using a combination of contrast loss and cross-entropy loss. In this paper, we detail the setup of hyperparameters:n-way, k-shot, β and the creation of support, query, and test datasets. Results Our method achieved over 98% accuracy on a query dataset with 35 samples per category using only 10 training samples per category. We documented the model’s loss, accuracy, and the confusion matrix of the results. Additionally, we employed the t-SNE algorithm to analyze and assess the model’s classification performance. Conclusion The proposed model demonstrates significant advantages in accuracy and minimal data dependency, performing robustly across all tested n-way, k-shot scenarios. It consistently achieved over 93% accuracy on comprehensive test datasets, including 1916 samples, confirming its high classification accuracy and strong generalization capability. Our research advances the use of few-shot learning in medical diagnostics, and also lays the groundwork for extending it to deal with rare, difficult-to-diagnose cases.

Deep Learning Driven Real-Time Airspace Monitoring Using Satellite Imagery

Detecting aircraft in remote sensing images poses a formidable challenge due to the diverse characteristics of aircraft, including type, size, pose, and intricate backgrounds. Traditional algorithms encounter difficulties in manually extracting features from numerous candidate regions. This paper introduces an innovative aircraft detection approach that combines corner clustering with a diverse set of Deep Learning (DL) models. The proposed method involves two main stages: region proposal and classification. In the region proposal stage, initial candidate regions are generated using a mean-shift clustering algorithm applied to corners detected on binary images. Subsequently, a comprehensive set of classifiers, encompassing CNN, DenseNet, MobileNetV2, Inception v3, Random Forest (R.F), ResNet50, ResNeXT, Support Vector Machine (SVM), VGG16, Xception, EfficientNet, and InceptionResNetv2, is employed for feature extraction and classification. The presented approach demonstrates superior accuracy and efficiency compared to conventional methods. By leveraging the autonomous learning capabilities of CNN and DL models on extensive datasets, the methodology generates a reduced yet high-quality set of candidate regions. Inspired by the detection methodology employed by image analysts, the approach adopts a coarse-to-fine strategy using CNN and DL models. The first CNN proposes coarse candidate regions, and the second identifies individual airplanes within these regions in finer detail. This framework results in a decreased number of candidate regions compared to existing literature while extracting distinctive deep features. Experimental evaluations on Google Earth images validate the efficiency of the proposed method, underscoring its potential for practical applications in both civilian and military contexts.

Enhancing forensic blood detection using hyperspectral imaging and advanced preprocessing techniques

Bloodstains are pivotal in forensic investigations as they can provide crucial DNA information about individuals involved in a crime. Traditional methods for bloodstain detection, including chemical tests and forensic lights, have limitations such as non-specificity to human blood and susceptibility to false positives. In forensic blood detection, molecular spectroscopy is crucial for identifying the unique spectral fingerprints of blood, which arise from its molecular composition. Hyperspectral imaging (HSI) leverages this principle by capturing a wide spectrum of light for each pixel in an image, allowing for the detailed analysis of various substances. HSI has emerged as a promising alternative, offering non-contact, rapid, and cost-effective detection of bloodstains by analyzing the visible and near-infrared electromagnetic spectrum. This study explores the application of HSI for blood detection, addressing challenges such as spectral mixing, time-related changes in bloodstain spectra, and data complexity. The study introduces a novel framework to optimize HSI data, enhancing the accuracy and efficiency of bloodstain classification, called the Fast Extraction (FE) framework. It includes two stages. The main method in the first one is the Enhancing Transformation Reduction (ETR) method to reduce the dimension and complexity of the HSI. The second stage contains a compatible classification model to enhance feature extraction and classification. Our approach is validated using the HyperBlood datasets and many evaluation methods, demonstrating superior performance compared to state-of-the-art deep learning models. It provides high accuracy (97%–100 %) for all HSIs, overcoming various difficulties and blood collection times.

EQID: Entangled quantum image descriptor an approach for early plant disease detection

In present day agriculture, early and accurate identification of plant diseases is essential for prompt response, which protects crop quality and output. This paper presents the Entangled Quantum-Inspired Deep learning model (EQID), a unique method that improves feature representation and classification in plant disease prediction by utilizing the concepts of quantum computing. Two different datasets with images of potatoes and tomatoes as leaves were used to test the EQID model, which performed better than traditional models. EQID obtained 98.96% accuracy, 98.98% precision, 98.96% recall, and 98.90% F1 score on images of potato leaves. For tomato leaves, comparable outcomes were noted, with accuracy, precision, recall, and F1 score all above 99.61%. The accuracy of disease prediction is greatly increased by the efficient and effective feature representation made possible by the EQID model's inclusion of quantum computing techniques. Additionally, the model outperformed other cutting-edge models such as DenseNet-121, VGGNet 16, and Xception Net, illustrating the potentially revolutionary effects of quantum-inspired models in agriculture. Future work will focus on applying the EQID model to a broader range of crops and plant diseases, as well as incorporating additional data sources to further enhance the model's predictive capabilities.

A Survey on Genetic Disease − Autism Spectrum Disorder Prediction and Classification in Machine Learning

Autism spectrum disorder (ASD) is a hereditary, neurological condition with many aetiologies that manifest in early childhood. Mental illnesses, including anxiety, poor communication, and a lack of recurrent interest, may result from ASD. It can be highly advantageous for children to improve their psychological wellness level if the ASD is recognized in the earlier years of life. Furthermore, machine learning (ML) approaches are now essential for diagnosing and categorising ASD. The creation of computer programmes that can acquire data and utilise it to gain knowledge for oneself is the main goal of this aspect of artificial intelligence. Many scholars have suggested various ML strategies for quickly and accurately detecting the various forms of ASD. This paper presents a survey on ASD prediction and classification using ML methods-based research articles from the year 2016 to 2023. Moreover, the current survey article discusses the performance assessment employing different metrics and made a comparative assessment to determine the ML model’s effectiveness. From this survey, it is identified that Artificial Neural Network model has attained better results than other ML algorithms. Moreover, further ASD studies employing an ML strategy for feature selection, prediction and classification can greatly benefit from this research.

Enhancing Autism Spectrum Disorder identification in multi-site MRI imaging: A multi-head cross-attention and multi-context approach for addressing variability in un-harmonized data

Multi-site MRI imaging poses a significant challenge due to the potential variations in images across different scanners at different sites. This variability can introduce ambiguity in further image analysis. Consequently, the image analysis techniques become site-dependent and scanner-dependent, implying that adjustments in the analysis methodologies are necessary for each scanner configuration. Further, implementing real-time modifications becomes intricate, particularly when incorporating a new type of scanner, as it requires adapting the analysis methods accordingly. Taking into account the aforementioned challenge, we have considered its implications for an Autism spectrum disorder (ASD) application. Our objective is to minimize the impact of site and scanner variability in the analysis, aiming to develop a model that remains effective across different scanners and sites.This entails devising a methodology that allows the same model to function seamlessly across multiple scanner configurations and sites. ASD, a behavioral disorder affecting child development, requires early detection. Clinical observation is time-consuming, prompting the use of fMRI with machine/deep learning for expedited diagnosis. Previous methods leverage fMRI’s functional connectivity but often rely on less generalized feature extractors and classifiers. Hence, there is significant room for improvement in the generalizability of detection methods across multi-site data, which is acquired from multiple scanners with different settings. In this study, we propose a Cross-Combination Multi-Scale Multi-Context Framework (CCMSMCF) capable of performing neuroimaging-based diagnostic classification of mental disorders for a multi-site dataset. Thus, this framework attains a degree of internal data harmonization, rendering it to some extent site and scanner-agnostic. Our proposed network, CCMSMCF, is constructed by integrating two sub-modules: the Multi-Head Attention Cross-Scale Module (MHACSM) and the Residual Multi-Context Module (RMCN). We also employ multiple loss functions in a novel manner for training the model, which includes Binary Cross Entropy, Dice loss, and Embedding Coupling loss. The model is validated on the Autism Brain Imaging Data Exchange I (ABIDE-I) dataset, which includes data from multiple scanners across different sites, and achieves promising results.

Multi-Scale Information Fusion and Decoupled Representation Learning for Robust Microbe-Disease Interaction Prediction

Research indicates that microbe activity within the human body significantly influences health by being closely linked to various diseases. Accurately predicting microbe-disease interactions (MDIs) offers critical insights for disease intervention and pharmaceutical research. Current advanced AI-based technologies automatically generate robust representations of microbes and diseases, enabling effective MDI predictions. However, these models continue to face significant challenges. A major issue is their reliance on complex feature extractors and classifiers, which substantially diminishes the models’ generalizability. To address this, we introduce a novel graph autoencoder framework that utilizes decoupled representation learning and multi-scale information fusion strategies to efficiently infer potential MDIs. Initially, we randomly mask portions of the input microbe-disease graph based on Bernoulli distribution to boost self-supervised training and minimize noise-related performance degradation. Secondly, we employ decoupled representation learning technology, compelling the graph neural network (GNN) to independently learn the weights for each feature subspace, thus enhancing its expressive power. Finally, we implement multi-scale information fusion technology to amalgamate the multi-layer outputs of GNN, reducing information loss due to occlusion. Extensive experiments on public datasets demonstrate that our model significantly surpasses existing top MDI prediction models. This indicates that our model can accurately predict unknown MDIs and is likely to aid in disease discovery and precision pharmaceutical research. Code and data are accessible at: https://github.com/shmildsj/MDI-IFDRL.

Research on 3D laser SLAM algorithm based on graph optimization

Abstract Aiming at the problems that the positioning accuracy of laser Simultaneous Localization and Mapping (SLAM) is affected by the misclassification of close feature points and incorrect feature matching in the environment with redundant features, a 3D laser SLAM algorithm based on graph optimization was proposed. Firstly, to reduce the feature extraction error problem, the distance threshold is used to filter the close point cloud and improve the accuracy of feature classification, enhanced reliability of optoelectronic measurement data. Secondly, to improve the consistency of feature matching, based on the rough matching based on k-dimensional tree (KD-Tree), a two-stage fine matching based on geometric consistency is introduced to improved laser point alignment accuracy. The proposed method is compared with the mainstream localization algorithms based on KITTI dataset. Compared with Lidar-IMU Odometry with Sparse Adaptive Mapping and Fast LiDAR-Inertial Odometry, the positioning error of the proposed method is reduced by 21.88%and 24.43%on average, which showsthe superiority of the present algorithm in the application of optoelectronics and optical technology, and provides new ideas and methods for the development of high-precision laser navigation technology.

A Sign Language Recognition using Improved Grey Wolf Optimization based neural networks

Sign language was developed to enable the deaf and hard of hearing community to communicate with society and convey information. It is the primary means by which they can interact with each other, as well as with the general population. The automatic gesture recognition system described in this paper uses Gray Wolf Optimization (GWO) to identify the most relevant features from a set of gesture images, with the aim of improving model performance. The main steps of the system are explained, including image preprocessing, feature extraction, GWO-based feature selection, classifier training, and evaluation. The proposed model achieved an accuracy of 99.9%.

An efficient approach for EMG controlled pattern recognition system based on MUAP identification and segregation

An Electromyography (EMG) based pattern recognition system constitutes various steps of signal processing and control engineering from signal acquisition to real-time control. Efficient control of external devices largely depends on the signal processing steps executed before the final output. This work presents a new approach to signal processing using Motor Unit Action Potential (MUAP) based signal decomposition and segmentation. An MUAP is a neurological response during muscle contraction. Due to the higher contact area of surface electrodes, MUAPs from multiple muscles are captured. An MUAP generated from a single muscle usually has identical waveshapes and similar discharging rates and usually lasts for 8–15 ms. These are known as primary MUAPs. The proposed algorithm identifies and uses the primary observed MUAPs for feature extraction and classification. Firstly, noise signals are eliminated by a determined noise margin, which also separates the active muscle movement signals. Next, a novel MUAP identification algorithm is implemented to detect the MUAP trains. Then, identified primary MUAPs are used to make segments with variable widths to extract feature vectors. Based on the correlation score of all the primary MUAPs, the segmentation is performed, which results in segmentation width varying from 110–200 ms. The achieved segmentation width is lesser than the conventional overlapping and non-overlapping methods — the proposed approach results in a 20 to 50% reduction in the segmentation width. Four different classifiers are tested during the machine learning stage to investigate the performance of the proposed approach. The obtained feature sets are then used to train the Linear Discriminant Analysis (LDA), K-Nearest Neighbor (kNN), Decision Tree (DT), and Random Forest (RF) classifiers. The classifiers are tested with precision, recall, F1 score, and accuracy. The kNN and DT classifiers performed better than the LDA and RF classifiers. The maximum precision and recall are 100% while the maximum achieved accuracy is 98.56%. The comparative results show higher accuracy even at lower segmentation widths than the conventional constant window scheme. The kNN and DT classifiers provide a 5% to 15% increment in accuracy compared to the constant window segmentation-based approach.

Multi-dimensional weighted deep subspace clustering with feature classification

Deep subspace clustering (DSC) methods based on deep autoencoder (DAE) and self-expression layers have achieved impressive performance. However, traditional DSC method often loses useful information in the feature extraction of DAE, leading to incomplete learning of the self-expression coefficient matrix. Besides, existing deep subspace clustering methods always extract feature information from a single dimension, ignoring the potential information that may exist in the self-expression coefficient matrix across multiple dimensions. In this paper, we propose a novel method called multi-dimensional weighted deep subspace clustering based on feature classification (MWDSC). Specifically, we use the multivariate feature classification module (MFC) to learn the potential information of multiple local feature matrix for local pre-clustering. To further improve the clustering performance, we propose a self-flip weighted fusion strategy (SWF). The SWF strategy obtains a self-expression coefficient matrix with rich features through self-flipping weighting, thereby aiding in achieving global optimal solutions. Extensive experiments conducted on five benchmark datasets validate the superiority and effectiveness of our proposed method over other state-of-the-art approaches.

Multi-modality artificial intelligence-based transthyretin amyloid cardiomyopathy detection in patients with severe aortic stenosis.

Transthyretin amyloid cardiomyopathy (ATTR-CM) is a frequent concomitant condition in patients with severe aortic stenosis (AS), yet it often remains undetected. This study aims to comprehensively evaluate artificial intelligence-based models developed based on preprocedural and routinely collected data to detect ATTR-CM in patients with severe AS planned for transcatheter aortic valve implantation (TAVI). In this prospective, single-center study, consecutive patients with AS were screened with [99mTc]-3,3-diphosphono-1,2-propanodicarboxylic acid ([99mTc]-DPD) for the presence of ATTR-CM. Clinical, laboratory, electrocardiogram, echocardiography, invasive measurements, 4-dimensional cardiac CT (4D-CCT) strain data, and CT-radiomic features were used for machine learning modeling of ATTR-CM detection and for outcome prediction. Feature selection and classifier algorithms were applied in single- and multi-modality classification scenarios. We split the dataset into training (70%) and testing (30%) samples. Performance was assessed using various metrics across 100 random seeds. Out of 263 patients with severe AS (57% males, age 83 ± 4.6years) enrolled, ATTR-CM was confirmed in 27 (10.3%). The lowest performances for detection of concomitant ATTR-CM were observed in invasive measurements and ECG data with area under the curve (AUC) < 0.68. Individual clinical, laboratory, interventional imaging, and CT-radiomics-based features showed moderate performances (AUC 0.70-0.76, sensitivity 0.79-0.82, specificity 0.63-0.72), echocardiography demonstrated good performance (AUC 0.79, sensitivity 0.80, specificity 0.78), and 4D-CT-strain showed the highest performance (AUC 0.85, sensitivity 0.90, specificity 0.74). The multi-modality model (AUC 0.84, sensitivity 0.87, specificity 0.76) did not outperform the model performance based on 4D-CT-strain only data (p-value > 0.05). The multi-modality model adequately discriminated low and high-risk individuals for all-cause mortality at a mean follow-up of 13 months. Artificial intelligence-based models using collected pre-TAVI evaluation data can effectively detect ATTR-CM in patients with severe AS, offering an alternative diagnostic strategy to scintigraphy and myocardial biopsy.

Automated Diabetic Retinopathy Detection Using Convolutional Neural Networks For Feature Extraction And Classification (ADRFEC)

Diabetic Retinopathy (DR) is a significant complication of Diabetes Mellitus, leading to various retinal abnormalities that can impair vision and, in severe cases, result in blindness. Approximately 80% of patients with long-standing diabetes for 10–15 years develop DR. The manual process of diagnosing and detecting DR for timely treatment is both time-consuming and unreliable, mainly due to resource constraints and the need for expert opinion. To address this challenge, computerized diagnostic systems utilizing Deep Learning (DL) Convolutional Neural Network (CNN) architectures have been proposed to learn DR patterns from fundus images and assess disease severity. The proposed model performs an exhaustive analysis of these architectures upon fundus images, and derives the best performing DL architecture for DR feature extraction and fundus image classification. Amongst all the models, ResNet50 has achieved the highest training accu- racy whereas VGG-16 has achieved the lowest training accuracy. Again, VGG-16 has achieved lowest validation accuracy whereas ResNet101 has achieved highest validation accuracy.

Electroencephalography Emotion Recognition Based on Rhythm Information Entropy Extraction

Electroencephalography (EEG) is a physiological signal directly generated by the central nervous system. Brain rhythm is closely related to a person’s emotional state and is widely used for EEG emotion recognition. In previous studies, the rhythm specificity between different brain channels was seldom explored. In this paper, the rhythm specificity of brain channels is studied to improve the accuracy of EEG emotion recognition. Variational mode decomposition is used to decompose rhythm signals and enhance features, and two kinds of information entropy, i.e., differential entropy (DE) and dispersion entropy (DispEn) are extracted. The rhythm being used to get the best result of single channel emotion recognition is selected as the representative rhythm, and the remove one method is employed to obtain rhythm information entropy feature. In the experiment, the DEAP database was used for EEG emotion recognition in valence-arousal space. The results showed that the best result of rhythm DE feature classification in the valence dimension is 77.04%, and the best result of rhythm DispEn feature classification in the arousal dimension is 79.25%.

Unconscious classification of quantitative electroencephalogram features from propofol versus propofol combined with etomidate anesthesia using one-dimensional convolutional neural network.

Establishing a convolutional neural network model for the recognition of characteristic raw electroencephalogram (EEG) signals is crucial for monitoring consciousness levels and guiding anesthetic drug administration. This trial was conducted from December 2023 to March 2024. A total of 40 surgery patients were randomly divided into either a propofol group (1% propofol injection, 10 mL: 100 mg) (P group) or a propofol-etomidate combination group (1% propofol injection, 10 mL: 100 mg, and 0.2% etomidate injection, 10 mL: 20 mg, mixed at a 2:1 volume ratio) (EP group). In the P group, target-controlled infusion (TCI) was employed for sedation induction, with an initial effect site concentration set at 5-6 μg/mL. The EP group received an intravenous push with a dosage of 0.2 mL/kg. Six consciousness-related EEG features were extracted from both groups and analyzed using four prediction models: support vector machine (SVM), Gaussian Naive Bayes (GNB), artificial neural network (ANN), and one-dimensional convolutional neural network (1D CNN). The performance of the models was evaluated based on accuracy, precision, recall, and F1-score. The power spectral density (94%) and alpha/beta ratio (72%) demonstrated higher accuracy as indicators for assessing consciousness. The classification accuracy of the 1D CNN model for anesthesia-induced unconsciousness (97%) surpassed that of the SVM (83%), GNB (81%), and ANN (83%) models, with a significance level of p < 0.05. Furthermore, the mean and mean difference ± standard error of the primary power values for the EP and P groups during the induced period were as follows: delta (23.85 and 16.79, 7.055 ± 0.817, p < 0.001), theta (10.74 and 8.743, 1.995 ± 0.7045, p < 0.02), and total power (24.31 and 19.72, 4.588 ± 0.7107, p < 0.001). Large slow-wave oscillations, power spectral density, and the alpha/beta ratio are effective indicators of changes in consciousness during intravenous anesthesia with a propofol-etomidate combination. These indicators can aid anesthesiologists in evaluating the depth of anesthesia and adjusting dosages accordingly. The 1D CNN model, which incorporates consciousness-related EEG features, represents a promising tool for assessing the depth of anesthesia. https://www.chictr.org.cn/index.html.

Uncovering tidal treasures: automated classification of faint tidal features in DECaLS data

ABSTRACT Tidal features are a key observable prediction of the hierarchical model of galaxy formation and contain a wealth of information about the properties and history of a galaxy. Modern wide-field surveys such as LSST and Euclid will revolutionize the study of tidal features. However, the volume of data will prohibit visual inspection to identify features, thereby motivating a need to develop automated detection methods. This paper presents a visual classification of ∼2000 galaxies from the DECaLS survey into different tidal feature categories: arms, streams, shells, and diffuse. We trained a convolutional neural network (CNN) to reproduce the assigned visual classifications using these labels. Evaluated on a testing set where galaxies with tidal features were outnumbered $\sim 1:10$, our network performed very well and retrieved a median $98.7\pm 0.3$, $99.1\pm 0.5$, $97.0\pm 0.8$, and $99.4^{+0.2}_{-0.6}$ per cent of the actual instances of arm, stream, shell, and diffuse features respectively for just 20 per cent contamination. A modified version that identified galaxies with any feature against those without achieved scores of $0.981^{+0.001}_{-0.003}$, $0.834^{+0.014}_{-0.026}$, $0.974^{+0.008}_{-0.004}$, and $0.900^{+0.073}_{-0.015}$ for the accuracy, precision, recall, and F1 metrics, respectively. We used a gradient-weighted class activation mapping analysis to highlight important regions on images for a given classification to verify the network was classifying the galaxies correctly. This is the first demonstration of using CNNs to classify tidal features into sub-categories, and it will pave the way for the identification of different categories of tidal features in the vast samples of galaxies that forthcoming wide-field surveys will deliver.

Research on Applying Deep Learning to Visual–Motor Integration Assessment Systems in Pediatric Rehabilitation Medicine

In pediatric rehabilitation medicine, manual assessment methods for visual–motor integration result in inconsistent scoring standards. To address these issues, incorporating artificial intelligence (AI) technology is a feasible approach that can reduce time and improve accuracy. Existing research on visual–motor integration scoring has proposed a framework based on convolutional neural networks (CNNs) for the Beery–Buktenica developmental test of visual–motor integration. However, as the number of training questions increases, the accuracy of this framework significantly decreases. This paper proposes a new architecture to reduce the number of features, channels, and overall model complexity. The architectureoptimizes input features by concatenating question numbers with answer features and selecting appropriate channel ratios and optimizes the output vector by designing the task as a multi-class classification. This paper also proposes a model named improved DenseNet. After experimentation, DenseNet201 was identified as the most suitable pre-trained model for this task and was used as the backbone architecture for improved DenseNet. Additionally, new fully connected layers were added for feature extraction and classification, allowing for specialized feature learning. The architecture can provide reasons for unscored results based on prediction results and decoding rules, offering directions for children’s training. The final experimental results show that the proposed new architecture improves the accuracy of scoring 6 question graphics by 12.8% and 12 question graphics by 20.14% compared to the most relevant literature. The accuracy of the proposed new architecture surpasses the model frameworks of the most relevant literature, demonstrating the effectiveness of this approach in improving scoring accuracy and stability.