Enhanced Feature Extraction Research Articles

Deformable multi-level feature network applied to nucleus segmentation

IntroductionThe nucleus plays a crucial role in medical diagnosis, and accurate nucleus segmentation is essential for disease assessment. However, existing methods have limitations in handling the diversity of nuclei and differences in staining conditions, restricting their practical application.MethodsA novel deformable multi-level feature network (DMFNet) is proposed for nucleus segmentation. This network is based on convolutional neural network and divides feature processing and mask generation into two levels. At the feature level, deformable convolution is used to enhance feature extraction ability, and multi-scale features are integrated through a balanced feature pyramid. At the mask level, a one-stage framework is adopted to directly perform instance segmentation based on location.ResultsExperimental results on the MoNuSeg 2018 dataset show that the mean average precision (mAP) and mean average recall (mAR) of DMFNet reach 37.8% and 47.4% respectively, outperforming many current advanced methods. Ablation experiments verified the effectiveness of each module of the network.DiscussionDMFNet provides an effective solution for nucleus segmentation and has important application value in medical image analysis.

A lightweight model for echo trace detection in echograms based on improved YOLOv8

With the rise of underwater unmanned platforms like unmanned boats, ROVs, and AUVs, there’s an increasing need for underwater detection technologies. Researchers have merged scientific echosounders with these platforms for biometric applications. However, current detection models are too parameter-heavy to embed in echosounders and struggle with noisy, irregular, and dense echograms. This paper introduces YOLOv8-SBE, a lightweight fish detection model based on YOLOv8, addressing these issues by enhancing feature extraction, information fusion, and small object recognition. YOLOv8-SBE adds the C2f_ScConv module to improve efficiency and reduce parameters, incorporates the BiFPN structure to enhance information transfer, and uses the EMA attention module for better small target recognition. It reduces computational complexity by 18.5%, decreases model parameters by 40%, and improves mAP0.5 to 79.5% and mAP0.5:0.95 to 58.2%, making it suitable for echosounders with limited resources.

Fundus Vascular Segmentation Based on Data Enhancement and Invariant Feature Extraction

Deep neural networks have emerged as the predominant method for medical image segmentation, owing to their robust feature learning capabilities, enabling accurate automatic segmentation of target structures within intricate medical images. Fundus images, being crucial in diagnosing ophthalmic diseases, underscore the importance of effective segmentation techniques. However, fundus vascular images pose challenges due to their high complexity and subtle individual differences, necessitating improvement in existing segmentation methodologies for enhanced disease classification accuracy. This paper introduces a fundus blood vessel segmentation model, employing data augmentation and invariant feature extraction, to systematically tackle the core challenges in medical image processing, particularly in fundus blood vessel segmentation. These challenges include limited source domain samples and inadequate model domain generalization. The model adopts a dual-dimensional strategy. Firstly, it delves into data augmentation technology to enhance the diversity and representativeness of samples within the finite source domain. This is achieved through an image enhancement module based on Fourier transform, mitigating the impact of data scarcity on model training effectiveness. Secondly, the research focuses on interdomain invariant feature extraction, aiming to extract feature representations that consistently characterize fundus blood vessel structure and pathology across different data distributions, thereby enhancing model generalization performance in unfamiliar domains. Specifically, the paper designs a fundus image enhancement module based on Fourier transform in the data augmentation dimension. In the feature extraction dimension, it proposes a normalization module based on uncertainty theory, departing from traditional normalization methods. Experimental results demonstrate the efficacy of the proposed method, showcasing superior generalization performance compared to existing techniques in retinal blood vessel and OD/OC segmentation tasks. Experience validates the model-agnostic nature of the learned strategy, indicating its potential for seamless transferability to other models, thereby offering robust support for advancing medical image segmentation research and applications.

Study of spot distance on resistance spot welding quality: a 1DCNN-BiLSTM-Attention-based online inspection method

Abstract Resistance spot welding (RSW) is widely employed in the automotive and home appliance industries due to its high efficiency, low cost, and suitability for automation. However, traditional quality detection methods rely on destructive testing, leading to inefficiencies and resource wastage. This paper presents a novel quality inspection model for RSW that utilizes a one-dimensional convolutional neural network, bidirectional long short-term memory network, and attention mechanism (1DCNN-BiLSTM-Attention) to address the challenges of extracting temporal data under varying spot distances. The model integrates a residual linking mechanism and Kolmogorov–Arnold Networks (KAN) to enhance feature extraction and performance. Experimental results reveal that the model demonstrates strong predictive capabilities across different spot distances, with particularly notable performance at 10 mm spacing, achieving a mean absolute error (MAE) of 0.0632, a root mean square error (RMSE) of 0.0603, and an R² value of 0.7513. These findings underscore the model’s ability to provide high-precision predictions, even under conditions influenced by significant shunt effects.

Rolling bearing fault diagnosis model based on external attention integrated convolutional neural network under imbalanced data conditions

Abstract Rolling bearings are essential components in numerous mechanical systems, and their failure can result in considerable downtime and expensive repairs. Therefore, accurate and timely fault diagnosis is vital for effective predictive maintenance and overall reliability. Traditional diagnostic methods often struggle with complex and non-stationary signals, compounded by issues of data imbalance in 
 realworld scenarios. A method for diagnosing rolling bearing faults has been developed in this paper utilizing External Attention (EA), Convolutional Neural Networks (CNN), and Continuous Wavelet Transform (CWT), specifically addressing the challenge of imbalanced sample data. This approach offers significant advantages, including a reduction in complexity by eliminating the need for data augmentation and leveraging external attention for enhanced feature extraction from samples. Compared to other attention mechanisms, this method demonstrates outstanding performance on both training and testing sets with imbalanced samples, exhibiting minimal overfitting tendencies. The proposed CWT-EACNN method effectively addresses the challenge of imbalanced sample data in rolling bearing fault diagnosis, demonstrating exceptional performance and reduced complexity.

EC-PFN: a multiscale woven fusion network for industrial product surface defect detection

In order to address challenges such as small target sizes, low contrast, significant intra-class variations, and indistinct inter-class differences in surface defect detection, this paper proposes the Enhanced Context-aware Parallel Fusion Network (EC-PFN). The network employs a Featur Weave Network architecture to enhance contextal awareess and parallel fusion capabilities. It utilizes a Feature Fusion Module (UniFusionLayer) for effective multiscale and multisemantic feature learning, offering new perspectives on feature fusion. Additionally, a Receptive Field Block (RFB) module is introduced to expand the receptive field, enhancing feature extraction in scenarios with low contrast and subtle defects. The Loss Ranking Module (LRM) is incorporated to optimize the target-oriented loss, improving performance by omitting low-confidence bounding boxes. Extensive experiments on a light guide plate defect dataset demonstrate that EC-PFN achieves a detection accuracy (mAP) of 98.9%, a detection speed of 92 FPS, and a computational cost of 14.5 GFLOPs, outperforming mainstream surface defect detection models.

DCFU‐Net: Rethinking an Effective Attention and Convolutional Architecture for Retinal Vessel Segmentation

ABSTRACTMorphological changes in retinal vessels are early indicators of cardiovascular and various fundus diseases. However, accurately segmenting thin blood vessels remains a challenge due to the complexity of the vascular structure and the irregularity of pathological features. This paper proposes a dual chain fusion U‐Net (DCFU‐Net) for the precise segmentation of retinal vessels. The network consists of a multi‐level segmentation network and a fusion network. The multi‐level segmentation network is designed with a dual chain architecture to generate segmentation results for both thick and thin vessels simultaneously. The fusion network combines the segmented thin and thick vessels with the original image, facilitating the generation of accurate segmentation outcomes. Notably, traditional convolution structures in the DCFU‐Net are replaced by dynamic snake convolutions (DS‐Conv). DS‐Conv is designed to adaptively focus on slender and tortuous local features, accurately capturing vascular structures. The shared weight residual block, integrating DS‐Conv and residual structures, which is called DS‐Res block. It serves as the backbone of the DCFU‐Net, enhancing feature extraction capabilities, while significantly reducing computational resource consumption. Additionally, this paper rethinks effective components of the Transformer architecture, identifying the inverted residual mobile block (IRMB) as a key element. By extending the DS‐Conv‐based IRMB into effective attention‐based (EAB) blocks, the network mitigates the loss of semantic information, thereby addressing inherent limitations. The DCFU‐Net is evaluated on three publicly available datasets: DRIVE, STARE, and CHASE_DB1. Qualitative and quantitative analyses demonstrate that the segmentation results of DCFU‐Net outperform state‐of‐the‐art methods.

A hybrid wind power prediction model based on seasonal feature decomposition and enhanced feature extraction

Efficient and accurate wind power prediction is crucial for enhancing the reliability and safety of power system. The data-driven forecasting methods are regarded as an effective solution. However, the inherent randomness and nonlinearity of wind power systems, along with the abundance of redundant information in measurement data, present challenges to forecasting methods. The integration of precise and efficient techniques for data feature decomposition and extraction is essential in conjunction with advanced data-driven forecasting models. Focus on the seasonal variation characteristics of wind energy, a hybrid wind power prediction model based on seasonal feature decomposition and enhanced feature extraction is proposed. The effectiveness and superiority of the proposed method in predictive accuracy are demonstrated through comprehensive multi-model experiment comparisons.

TMSA-Net:A novel attention mechanism for improved motor imagery EEG signal processing

Electroencephalography (EEG) is a non-invasive method used to record the brain’s electrical activity, widely employed in brain-computer interface (BCI) applications for decoding motor imagery (MI) signals. Traditional models combining Convolutional Neural Networks (CNNs) and Transformers for decoding Motor Imagery Electroencephalography (MI-EEG) signals often struggle to capture the crucial interrelationships between local and global features effectively, resulting in suboptimal performance. To address this issue, we propose a novel computationally efficient model that integrates CNNs with a novel attention mechanism within the Transformer architecture. This approach effectively captures both local and global dependencies, thereby enhancing feature extraction and decoding accuracy. Evaluations against state-of-the-art models (EEGNet, Deep ConvNet, IFNet, Conformer) on three public datasets (BCIC-IV-2a, BCIC-IV-2b, HGD) reveal substantial performance enhancements. Specifically, our model achieves improvements in accuracy of 9.78%, 11.05%, 4.81%, and 3.75% over these models on the BCIC-IV-2a dataset. Ablation studies and visualization techniques (t-SNE, Grad-CAM) further corroborate the effectiveness and interpretability of our model, highlighting its superior decoding capabilities. The code is available at https://github.com/Whit3Zhao/TMSA-Net.git.

Dynamic RU-Next: Advancing liver and tumor segmentation with enhanced deep learning architecture

Liver cancer is a significant worldwide health issue, making accurate liver tumor identification essential for diagnosis and treatment planning. However, achieving precise tumor segmentation in CT scans is complex due to structural, boundary, and density variations. The Dynamic RU-Next (DRuN) model is introduced to address these challenges, integrating a ResNeXt backbone for enhanced feature extraction with a Dynamic U-Net decoder utilizing pixel shuffle up-sampling. Two variants are tested on the publicly available LiTS dataset: DRuN-50 (ResNeXt-50 backbone) and DRuN-101 (ResNeXt-101 backbone). Both models demonstrated improved performance compared to existing methods, with DRuN-50 and DRuN-101 achieving accuracies of 98.33% and 98.88% for liver delineation, and 96.66% for tumor delineation. The respective dice scores are 98.41% and 98.96% for the liver and 96.29% and 95.38% for tumor segmentation. Additionally, these models process each image in 1.036 s and 1.075 s, respectively.

Fast Quality Detection of Astragalus Slices Using FA-SD-YOLO

Quality inspection is a pivotal component in the intelligent sorting of Astragalus membranaceus (Huangqi), a medicinal plant of significant pharmacological importance. To improve the precision and efficiency of assessing the quality of Astragalus slices, we present the FA-SD-YOLO model, an innovative advancement over the YOLOv8n architecture. This model introduces several novel modifications to enhance feature extraction and fusion while reducing computational complexity. The FA-SD-YOLO model replaces the conventional C2f module with the C2F-F module, developed using the FasterNet architecture, and substitutes the SPPF module with the Adaptive Inverted Fusion (AIFI) module. These changes markedly enhance the model’s feature fusion capabilities. Additionally, the integration of the SD module into the detection head optimizes parameter efficiency while improving detection performance. Performance evaluation highlights the superiority of the FA-SD-YOLO model. It achieves accuracy and recall rates of 88.6% and 89.6%, outperforming the YOLOv8n model by 1.8% and 1.3%, respectively. The model’s F1 score reaches 89.1%, and the mean average precision (mAP) improves to 93.2%, reflecting increases of 1.6% and 2.4% over YOLOv8n. These enhancements are accompanied by significant reductions in model size and computational cost: the parameter count is reduced to 1.58 million (a 47.3% reduction), and the FLOPS drops to 4.6 G (a 43.2% reduction). When compared with other state-of-the-art models, including YOLOv5s, YOLOv6s, YOLOv9t, and YOLOv11n, the FA-SD-YOLO model demonstrates superior performance across key metrics such as accuracy, F1 score, mAP, and FLOPS. Notably, it achieves a remarkable recognition speed of 13.8 ms per image, underscoring its efficiency and suitability for real-time applications. The FA-SD-YOLO model represents a robust and effective solution for the quality inspection of Astragalus membranaceus slices, providing reliable technical support for intelligent sorting machinery in the processing of this important medicinal herb.

Lightweight Mulberry Fruit Detection Method Based on Improved YOLOv8n for Automated Harvesting

Aiming at the difficulty of feature extraction in complex environments during mulberry detection and the need for embedded devices to lighten the model, this study carries out lightweight improvements on the basis of the YOLOv8n model. First, the CSPPC module incorporates lightweight partial convolution (PConv) within its bottleneck structure, replacing the C2f module to enhance feature extraction efficiency. Secondly, the ADown module is used to replace the traditional downsampling module and the P-Head module is used to replace the traditional convolutional detector head with the partial convolutional detector head. Finally, a knowledge distillation technique is used to compensate for the loss of accuracy due to parameter reduction. Ablation experiments are conducted to evaluate the impact of each module on the model’s performance. The experimental results show that the improved YOLOv8 model has a precision of 88.9%, a recall of 78.1%, and an average precision of 86.8%. The number of parameters is 1.29 × 106, the model size is 2.6 MB, the floating-point computation is 2.6 GFLOPs, and the frame rate reaches 19.84 FPS on the edge end. Therefore, this model provides theoretical and technical support for the deployment and application of mobile detection devices, such as automatic mulberry harvesting in practical scenarios.

A Lightweight Transmission Line Foreign Object Detection Algorithm Incorporating Adaptive Weight Pooling

Aerial photography using unmanned aerial vehicles (UAVs) to detect foreign objects is an important method to ensure the safety of transmission lines. However, existing detection algorithms often encounter challenges in complex environments, including limited recognition capability and high computational demands. To address these issues, this paper proposes YOLO-LAF, a lightweight foreign object detection algorithm that is based on YOLOv8n and incorporates an innovative adaptive weight pooling technique. The proposed method introduces a novel adaptive weight pooling module within the backbone network to enhance feature extraction for detecting foreign objects on transmission lines. Additionally, a multi-scale detection strategy is designed to integrate the FasterBlock and EMA modules. This combination enables the model to effectively capture both global and local image features through cross-channel interactions, thereby reducing misdetection and omission rates. Furthermore, a C2f-SCConv module is introduced in the neck network to streamline the model by eliminating redundant features, thus improving computational efficiency. Experimental results demonstrate that YOLO-LAF achieves average accuracies of 91.2% and 85.3% on the Southern Power Grid and RailFOD23 datasets, respectively, outperforming the original YOLOv8n algorithm by 2.6% and 1.8%. Moreover, YOLO-LAF reduces the number of parameters by 23.5% and 14.8% and the computational costs by 19.9% and 24.8%, respectively. These improvements demonstrate the superior detection performance of YOLO-LAF compared to other mainstream detection algorithms.

DDANet: A deep dilated attention network for intracerebral haemorrhage segmentation.

Intracranial haemorrhage (ICH) is an urgent and potentially fatal medical condition caused by brain blood vessel rupture, leading to blood accumulation in the brain tissue. Due to the pressure and damage it causes to brain tissue, ICH results in severe neurological impairment or even death. Recently, deep neural networks have been widely applied to enhance the speed and precision of ICH detection yet they are still challenged by small or subtle hemorrhages. The authors introduce DDANet, a novel haematoma segmentation model for brain CT images. Specifically, a dilated convolution pooling block is introduced in the intermediate layers of the encoder to enhance feature extraction capabilities of middle layers. Additionally, the authors incorporate a self-attention mechanism to capture global semantic information of high-level features to guide the extraction and processing of low-level features, thereby enhancing the model's understanding of the overall structure while maintaining details. DDANetalso integrates residual networks, channel attention, and spatial attention mechanisms for joint optimisation, effectively mitigating the severe class imbalance problem and aiding the training process. Experiments show that DDANet outperforms current methods, achieving the Dice coefficient, Jaccard index, sensitivity, accuracy, and a specificity of 0.712, 0.601, 0.73, 0.997, and 0.998, respectively. The code is available at https://github.com/hpguo1982/DDANet.

Research on Deep Learning Model Enhancements for PCB Surface Defect Detection

With the miniaturization and increasing complexity of electronic devices, the accuracy and efficiency of printed circuit board (PCB) defect detection are crucial to ensuring product quality. To address the issues of small defect sizes and high missed detection rates in PCB surface inspection, this paper proposes an enhanced YOLOv8s model which not only improves detection performance but also achieves a lightweight design. Firstly, the Nexus Attention module is introduced, which organically integrates multiple attention mechanisms to further enhance feature extraction and fusion capabilities, improving the model’s learning and generalization performance. Secondly, an improved CGFPN network is designed to optimize multi-scale feature fusion, significantly boosting the detection of small objects. Additionally, the WaveletUnPool module is incorporated, leveraging wavelet transform technology to refine the upsampling process, accurately restoring detailed information and improving small-object detection in complex backgrounds. Lastly, the C2f-GDConv module replaces the traditional C2f module, reducing the number of model parameters and computational complexity while maintaining feature extraction efficiency. Comparative experiments on a public PCB dataset demonstrate that the enhanced model achieved a mean average precision (mAP) of 97.3% in PCB defect detection tasks, representing a 3.0% improvement over the original model, while reducing Giga Floating Point Operations (GFLOPs) by 26.8%. These enhancements make the model more practical and adaptable for industrial applications, providing a solid foundation for future research.

Enhancing Data Privacy Protection and Feature Extraction in Secure Computing Using a Hash Tree and Skip Attention Mechanism

This paper addresses the critical challenge of secure computing in the context of deep learning, focusing on the pressing need for effective data privacy protection during transmission and storage, particularly in sensitive fields such as finance and healthcare. To tackle this issue, we propose a novel deep learning model that integrates a hash tree structure with a skip attention mechanism. The hash tree is employed to ensure data integrity and security, enabling the rapid verification of data changes, while the skip attention mechanism enhances computational efficiency by allowing the model to selectively focus on important features, thus minimizing unnecessary processing. The primary objective of our research is to develop a secure computing model that not only safeguards data privacy but also optimizes feature extraction capabilities. Our experimental results on the CIFAR-10 dataset demonstrate significant improvements over traditional models, achieving a precision of 0.94, a recall of 0.89, an accuracy of 0.92, and an F1-score of 0.91, notably outperforming standard self-attention and CBAM. Additionally, the visualization of results confirms that our approach effectively balances efficient feature extraction with robust data privacy protection. This research contributes a new framework for secure computing, addressing both the security and efficiency concerns prevalent in current methodologies.

Single-Task Joint Learning Model for an Online Multi-Object Tracking Framework

Multi-object tracking faces critical challenges, including occlusions, ID switches, and erroneous detection boxes, which significantly hinder tracking accuracy in complex environments. To address these issues, this study proposes a single-task joint learning (STJL) model integrated into an online multi-object tracking framework to enhance feature extraction and model robustness across diverse scenarios. Employing cross-dataset training, the model has improved generalization capabilities and can effectively handle various tracking conditions. A key innovation is the refined tracker initialization strategy that combines detection and tracklet confidence, which significantly reduces the number of false positives and ID switches. Additionally, the framework employs a combination of Mahalanobis and cosine distances to optimize data association, further improving tracking accuracy. The experimental results demonstrate that the proposed model outperformed state-of-the-art methods on standard benchmark datasets, achieving superior MOTA and reduced ID switches, confirming its effectiveness in dynamic and occlusion-heavy environments.

A computationally efficient method for induction motor bearing fault detection based on parallel convolutions and semi-supervised GAN

ABSTRACT Accurate and timely bearing fault detection is imperative for optimal system functioning and the implementation of preventative maintenance measures. Deep learning models provide viable solutions to these malfunctions, however, the lack of labelled data makes the training both expensive and cumbersome. To remedy this, various semi-supervised approaches have surfaced in the last decade, significantly mitigating the need for extensive labelled data but with added computational cost. This study proposes one such approach by leveraging generative adversarial networks (GAN) trained on a time-frequency based representation. The proposed Parallel Convolutions Semi-Supervised GAN, namely PC-SSGAN, uses bottleneck parallel convolutions blocks to capture multi-scale features in both local and global contexts, lacing both the generator and discriminator with enhanced feature extraction capabilities and simultaneously reducing the parameters and training time. The Proposed framework is evaluated on two distinct open-source datasets. The classification accuracy for both models exceeded 99.50%. Moreover, the proposed parallel convolutions-based architecture spent approximately 33% less time on training than the normal convolutional layers. It has been foreseen that the proposed fault detection system can be integrated into the motor fault tolerant control system to produce a unified framework that can make informed decisions to handle the bearing faults effectively.

Weak ultrasonic guided wave signal recognition based on one-dimensional convolutional neural network denoising autoencoder and its application to small defect detection in pipelines

Pipeline failures are often caused by the expansion of small defects. Structural damage to pipelines can lead to major safety accidents. When ultrasonic guided wave (UGW) technology is used for pipeline failure detection, the echoes produced by small defects manifest as weak UGW signals amidst significant noise. The low amplitude of these signals or complete drowning by noise makes them difficult to recognize. This study innovatively introduces a one-dimensional convolutional neural network denoising autoencoder (1DCNN-based DAE) for noise reduction in UGW signals using deep learning. To improve the conventional DAE, the model incorporated the Parametric Rectified Linear Unit (PReLU) activation function and a CNN for enhanced feature extraction, resulting in the proposed 1DCNN-based DAE. The model is trained on an extensive dataset of mixed signals with strong noise and their corresponding clean signals, enabling autonomous denoising in an unsupervised manner. Additionally, this paper proposes the application of the window-shifted power spectrum method for analyzing the denoised signals to identify and locate pipeline defects. The method involves traversing the signal with a window to intercept fragments, calculating their power, and plotting the power spectrum curve. Defects are then located based on the peak positions of this curve. Numerical simulation and experimental signals were used to validate the proposed method. Simulation results showed that the proposed 1DCNN-based DAE effectively improved the signal-to-noise ratio (SNR) of UGW mixed signals from −9 dB to 21.63 dB, representing an improvement of up to 30.63 dB. Experimental results demonstrated that the method accurately detected weak UGW signals from small defective pipes with a 2 % cross-section loss rate, achieving over 90 % recognition confidence and less than 1.5 % axial positioning error rate. In summary, the proposed 1DCNN-based DAE can effectively improve the SNR of the signal, reduce the noise in the UGW detection signal, and improve the sensitivity of defect identification; the window-shifted power spectrum method has a advantage in the accurate localization of defects.

RJ-TinyViT: an efficient vision transformer for red jujube defect classification

Compared to the surface defect detection of industrial products produced according to specified processes, the detection of surface defects in naturally grown red jujubes poses unique and significant challenges for researchers. The high diversity of surface defects, subtle distinctions from the background, low contrast, varying scales, and the presence of high levels of noise in images are among the factors that greatly amplify the complexity of defect detection tasks. Existing methods show some deficiencies in addressing these issues, mainly due to insufficient feature extraction capabilities and overly complex network structures, leading to limitations in model efficiency and practical application performance. To tackle the challenges associated with red jujube surface defect detection, this study proposes an optimized Tiny Vision Transformer (TinyViT) network structure, named RJ-TinyViT. This method refines the TinyViT-5 m network structure to reduce network burden and introduces an improved Multi-Kernel Block (MK Block) and an improved Mobile Inverted Bottleneck Convolution Block (MBConv Block) to enhance feature extraction capabilities. Additionally, we have integrated the Coordinate Attention (CA) module to enhance the model’s capacity for recognizing and focusing on features of surface defects on red jujubes. Experimental results show that RJ-TinyViT achieved a classification accuracy of 93.38%, marking an improvement of 1.84% over the original TinyViT network. At the same time, its Floating-point Operations (FLOPs) and Parameters (Params) were reduced to 58.97% and 39.84% of the original TinyViT network, respectively. These results not only demonstrate that RJ-TinyViT achieves model lightweighting while maintaining high accuracy but also highlight its value in practical industrial applications.