Bin Qin
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Bin Qin has a total of 50 co-authors. Here are the top co-authors by number of shared publications:
- Shang Gao
School of Computer, Jiangsu University of Science and Technology, Zhenjiang, China
7 shared publications - Shitong Wang
School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, China
5 shared publications - Hualong Yu
School of Computer, Jiangsu University of Science and Technology, Zhenjiang 212003, China
4 shared publications - Changbin Shao
School of Computer, Jiangsu University of Science and Technology, Zhenjiang, China
3 shared publications - Wei Li
School of Computer, Jiangsu University of Science and Technology, Zhenjiang 212003, China
3 shared publications - Yuhang Xie
School of Computer, Jiangsu University of Science and Technology, Zhenjiang 212003, China
3 shared publications
Latest Publications
- Open Access
Online Adaptive TSK Fuzzy Classifier with Incremental–Decremental Update Scheme for Non-Stationary Data Streams
Takagi–Sugeno–Kang (TSK) fuzzy classifiers have achieved notable success in pattern recognition and data mining due to their combination of approximation capability and model interpretability. However, existing TSK classifiers predominantly rely on batch training, which becomes impractical in streaming environments characterized by continuous sample arrival, limited memory, and evolving data distributions. To address these challenges, this paper proposes the Online Adaptive TSK (OA-TSK) fuzzy classifier with an incremental–decremental update protocol for non-stationary data streams. First, we develop the Decremental Adaptive TSK Fuzzy Classifier (DA-TSK), whose update procedure applies the Sherman–Morrison formula to a symmetric rank-one downdate, enabling precise sample removal while maintaining exact equivalence to batch least-squares retraining. Building upon DA-TSK and our prior Dynamic Incremental Adaptive TSK Fuzzy Classifier (DIA-TSK), we then propose OA-TSK, which integrates incremental and decremental updates within a sliding-window mechanism to achieve continuous adaptation with bounded memory. The key theoretical contribution is that, in exact arithmetic, OA-TSK produces at every time step the same ridge-regularized least-squares solution as batch retraining on the active window. To further handle severe shifts in the input distribution, which can cause fixed rule centers to become poorly aligned with newly arriving samples, we propose Online Adaptive TSK with Antecedent Reclustering (OA-A-TSK), which incorporates an adaptive antecedent reclustering mechanism. Comprehensive experiments on synthetic and real-world data streams demonstrate that OA-TSK achieves classification accuracy comparable to that of ridge-regularized batch retraining while substantially reducing per-sample computational complexity. An analysis of average ranks shows that OA-TSK and OA-A-TSK rank among the top-performing methods across diverse drift scenarios while preserving the intrinsic interpretability of TSK fuzzy systems.
Open AccessOnline Adaptive TSK Fuzzy Classifier with Incremental–Decremental Update Scheme for Non-Stationary Data Streams
Takagi–Sugeno–Kang (TSK) fuzzy classifiers have achieved notable success in pattern recognition and data mining due to their combination of approximation capability and model interpretability. However, existing TSK classifiers predominantly rely on batch training, which becomes impractical in streaming environments characterized by continuous sample arrival, limited memory, and evolving data distributions. To address these challenges, this paper proposes the Online Adaptive TSK (OA-TSK) fuzzy classifier with an incremental–decremental update protocol for non-stationary data streams. First, we develop the Decremental Adaptive TSK Fuzzy Classifier (DA-TSK), whose update procedure applies the Sherman–Morrison formula to a symmetric rank-one downdate, enabling precise sample removal while maintaining exact equivalence to batch least-squares retraining. Building upon DA-TSK and our prior Dynamic Incremental Adaptive TSK Fuzzy Classifier (DIA-TSK), we then propose OA-TSK, which integrates incremental and decremental updates within a sliding-window mechanism to achieve continuous adaptation with bounded memory. The key theoretical contribution is that, in exact arithmetic, OA-TSK produces at every time step the same ridge-regularized least-squares solution as batch retraining on the active window. To further handle severe shifts in the input distribution, which can cause fixed rule centers to become poorly aligned with newly arriving samples, we propose Online Adaptive TSK with Antecedent Reclustering (OA-A-TSK), which incorporates an adaptive antecedent reclustering mechanism. Comprehensive experiments on synthetic and real-world data streams demonstrate that OA-TSK achieves classification accuracy comparable to that of ridge-regularized batch retraining while substantially reducing per-sample computational complexity. An analysis of average ranks shows that OA-TSK and OA-A-TSK rank among the top-performing methods across diverse drift scenarios while preserving the intrinsic interpretability of TSK fuzzy systems.
- Open Access
- 31 July 2026
- Published by MDPI AG in Fractal and Fractional
- Vol. 10 (8) , 526
Fractional-Order Memory and Elite-Center-Guided Sine–Cosine Optimization for Feature Selection
Metaheuristic optimization algorithms are widely used in wrapper-based feature selection, since they can search complex combinatorial spaces without gradient information. The standard sine cosine algorithm (SCA) is simple and easy to implement, but depends on the current population state and a single global optimal guide. This dependence can cause premature convergence and late-stage oscillations in discretized feature-selection tasks. To address these issues, we propose the fractional-order elite-memory sine cosine algorithm (FOSCA). The FOSCA integrates short-memory fractional-order position reconstruction, dynamic elite-center guidance and nonlinear search-factor decay into the SCA search dynamics. These mechanisms improve the trajectory continuity, guidance diversity and convergence stability. Experiments on 14 classification datasets showed that the FOSCA achieved a competitive accuracy, F1-score, precision and recall. The FOSCA also outperformed the original SCA on 12 out of the 14 datasets. Statistical tests, a stability analysis, a performance–sparsity trade-off analysis and ablation studies confirmed the effectiveness of the proposed mechanisms. Overall, the FOSCA improves the SCA search reliability in discrete feature selection and offers a reproducible basis for related optimizer-based methods.
- 31 July 2026
- Published by MDPI AG in Fractal and Fractional
- Vol. 10 (8) , 526
Open AccessFractional-Order Memory and Elite-Center-Guided Sine–Cosine Optimization for Feature Selection
Metaheuristic optimization algorithms are widely used in wrapper-based feature selection, since they can search complex combinatorial spaces without gradient information. The standard sine cosine algorithm (SCA) is simple and easy to implement, but depends on the current population state and a single global optimal guide. This dependence can cause premature convergence and late-stage oscillations in discretized feature-selection tasks. To address these issues, we propose the fractional-order elite-memory sine cosine algorithm (FOSCA). The FOSCA integrates short-memory fractional-order position reconstruction, dynamic elite-center guidance and nonlinear search-factor decay into the SCA search dynamics. These mechanisms improve the trajectory continuity, guidance diversity and convergence stability. Experiments on 14 classification datasets showed that the FOSCA achieved a competitive accuracy, F1-score, precision and recall. The FOSCA also outperformed the original SCA on 12 out of the 14 datasets. Statistical tests, a stability analysis, a performance–sparsity trade-off analysis and ablation studies confirmed the effectiveness of the proposed mechanisms. Overall, the FOSCA improves the SCA search reliability in discrete feature selection and offers a reproducible basis for related optimizer-based methods.
- Open Access
Self-Adaptive AdamW-Guided Optimization: A Learning-Driven Metaheuristic for Solving Complex Real-World Engineering Problems
Given the growing complexity of continuous optimization problems in strongly coupled and black-box environments, this study proposes a novel adaptive gradient-guided metaheuristic, referred to as Self-Adaptive AdamW-Guided Optimization (SAWG). Without requiring explicit gradient information, SAWG constructs population-based pseudo-gradients and systematically integrates key AdamW mechanisms, including adaptive moment estimation, step-size regulation, and weight decay, to guide efficient population updates. Furthermore, a stagnation-aware adaptive control strategy is introduced to alleviate premature convergence and dynamically balance exploration and exploitation. To evaluate the optimization performance of SAWG, experiments were conducted on the CEC2017 and CEC2020 benchmark suites and eight engineering optimization problems. SAWG was also compared with nine other typical and novel high-performance optimizers. Experimental results and statistical analysis show that SAWG achieved excellent optimization performance in most test tasks and maintained strong adaptability and competitiveness in various numerical optimization problems. Therefore, SAWG can be regarded as a high-performance optimizer, providing a novel and effective method for solving complex numerical optimization tasks.
Open AccessSelf-Adaptive AdamW-Guided Optimization: A Learning-Driven Metaheuristic for Solving Complex Real-World Engineering Problems
Given the growing complexity of continuous optimization problems in strongly coupled and black-box environments, this study proposes a novel adaptive gradient-guided metaheuristic, referred to as Self-Adaptive AdamW-Guided Optimization (SAWG). Without requiring explicit gradient information, SAWG constructs population-based pseudo-gradients and systematically integrates key AdamW mechanisms, including adaptive moment estimation, step-size regulation, and weight decay, to guide efficient population updates. Furthermore, a stagnation-aware adaptive control strategy is introduced to alleviate premature convergence and dynamically balance exploration and exploitation. To evaluate the optimization performance of SAWG, experiments were conducted on the CEC2017 and CEC2020 benchmark suites and eight engineering optimization problems. SAWG was also compared with nine other typical and novel high-performance optimizers. Experimental results and statistical analysis show that SAWG achieved excellent optimization performance in most test tasks and maintained strong adaptability and competitiveness in various numerical optimization problems. Therefore, SAWG can be regarded as a high-performance optimizer, providing a novel and effective method for solving complex numerical optimization tasks.
- Open Access
- 17 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. PP, 1
SUA-YOLO: A Novel YOLO-based Algorithm for Detecting Dense Small Objects from a Low-altitude UAV Perspective
Unmanned aerial vehicle (UAV) aerial photography technology is crucial in various fields, including daily life, production, and security. However, it faces challenges in small object detection, such as extremely small objects, large scale variations, dense occlusions, and complex backgrounds. This paper presents SUA-YOLO, an improved solution based on the YOLOv8 model, to tackle these issues. First, a fine-grained detection layer is added to reconstruct the model framework, addressing the feature compression problem in deep networks. Second, the improved C2f_SCSAR module embeds Spatial and Channel Synergistic Attention (SCSA) to suppress background noise and enhance critical feature responses, thereby improving the model’s feature extraction capability in complex backgrounds. Next, the RepNCSPELAN4 module employs a multi-branch architecture to enhance feature representation, thereby improving cross-scale feature fusion of the Neck. Additionally, the Universal Adaptive Feature Fusion (UAFF) mechanism is constructed to balance detail and semantic features, further enhancing feature fusion capabilities and improving the model’s robustness across various scenarios. Finally, the Inner-WIoU loss function is integrated to improve the model’s adaptability in dense occlusion situations, effectively addressing false negatives. Using YOLOv8n as the baseline, we conducted object detection experiments on the VisDrone2019 dataset. The improved model achieved increases of 7.1% in mAP@0.5 and 4.8% in mAP@0.5:0.95. Additionally, generalization experiments on the DOTA dataset showed that SUA-YOLO performs effectively in small object detection tasks across various scenarios.
- 17 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. PP, 1
Open AccessSUA-YOLO: A Novel YOLO-based Algorithm for Detecting Dense Small Objects from a Low-altitude UAV Perspective
Unmanned aerial vehicle (UAV) aerial photography technology is crucial in various fields, including daily life, production, and security. However, it faces challenges in small object detection, such as extremely small objects, large scale variations, dense occlusions, and complex backgrounds. This paper presents SUA-YOLO, an improved solution based on the YOLOv8 model, to tackle these issues. First, a fine-grained detection layer is added to reconstruct the model framework, addressing the feature compression problem in deep networks. Second, the improved C2f_SCSAR module embeds Spatial and Channel Synergistic Attention (SCSA) to suppress background noise and enhance critical feature responses, thereby improving the model’s feature extraction capability in complex backgrounds. Next, the RepNCSPELAN4 module employs a multi-branch architecture to enhance feature representation, thereby improving cross-scale feature fusion of the Neck. Additionally, the Universal Adaptive Feature Fusion (UAFF) mechanism is constructed to balance detail and semantic features, further enhancing feature fusion capabilities and improving the model’s robustness across various scenarios. Finally, the Inner-WIoU loss function is integrated to improve the model’s adaptability in dense occlusion situations, effectively addressing false negatives. Using YOLOv8n as the baseline, we conducted object detection experiments on the VisDrone2019 dataset. The improved model achieved increases of 7.1% in mAP@0.5 and 4.8% in mAP@0.5:0.95. Additionally, generalization experiments on the DOTA dataset showed that SUA-YOLO performs effectively in small object detection tasks across various scenarios.
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 105314-105327
Improved YOLOv8 Algorithm for Small Object Detection in UAV Aerial Images
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 105314-105327
Open AccessImproved YOLOv8 Algorithm for Small Object Detection in UAV Aerial Images
- Open Access
An Enhanced Plant Growth Algorithm with Adam Learning, Lévy Flight, and Dynamic Stage Control
This study addresses the limitations of the traditional Plant Growth Algorithm (PGA), including insufficient local exploitation, premature convergence, and performance degradation in high-dimensional optimization. To enhance search efficiency, we propose the ALDPGA (Adam–Lévy Dynamic Plant Growth Algorithm), which incorporates Adam-based adaptive gradient learning, Lévy long-tailed perturbation, and a dynamic stage-control mechanism. The method strengthens directional refinement in the light region using gradient-assisted learning and a simulated-annealing rule, while staged hybrid perturbations and adaptive learning-rate scheduling expand early exploration in the shaded region. During the cell-elongation phase, Lévy-driven dynamic trajectories guide the transition from global search to fine-grained convergence. Notably, the light and shaded regions of the algorithm are designed symmetrically, balancing exploration and exploitation. The light region reflects phototropism, fostering growth towards optimal solutions, while the shaded region adapts to explore previously underexplored areas. Extensive experiments on CEC2017, CEC2020, and CEC2022 benchmarks demonstrate improvements in optimal solutions, convergence speed, and statistical stability. Wilcoxon tests confirm the significance of these gains, and ablation studies verify the contributions of each component. ALDPGA’s enhanced robustness and optimization efficiency make it well suited for complex, multimodal, and high-dimensional problems, offering new insights into bio-inspired optimization frameworks.
Open AccessAn Enhanced Plant Growth Algorithm with Adam Learning, Lévy Flight, and Dynamic Stage Control
This study addresses the limitations of the traditional Plant Growth Algorithm (PGA), including insufficient local exploitation, premature convergence, and performance degradation in high-dimensional optimization. To enhance search efficiency, we propose the ALDPGA (Adam–Lévy Dynamic Plant Growth Algorithm), which incorporates Adam-based adaptive gradient learning, Lévy long-tailed perturbation, and a dynamic stage-control mechanism. The method strengthens directional refinement in the light region using gradient-assisted learning and a simulated-annealing rule, while staged hybrid perturbations and adaptive learning-rate scheduling expand early exploration in the shaded region. During the cell-elongation phase, Lévy-driven dynamic trajectories guide the transition from global search to fine-grained convergence. Notably, the light and shaded regions of the algorithm are designed symmetrically, balancing exploration and exploitation. The light region reflects phototropism, fostering growth towards optimal solutions, while the shaded region adapts to explore previously underexplored areas. Extensive experiments on CEC2017, CEC2020, and CEC2022 benchmarks demonstrate improvements in optimal solutions, convergence speed, and statistical stability. Wilcoxon tests confirm the significance of these gains, and ablation studies verify the contributions of each component. ALDPGA’s enhanced robustness and optimization efficiency make it well suited for complex, multimodal, and high-dimensional problems, offering new insights into bio-inspired optimization frameworks.
- Open Access
- 28 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 133483-133496
LSD-Det: A Lightweight Detector for Small Ship Targets in SAR Images
Synthetic Aperture Radar (SAR) plays a vital role in ship safety monitoring and marine environmental protection. However, ship targets in SAR images are often small, have blurred edges, and suffer from strong background interference, making it difficult for traditional detection algorithms to balance accuracy and real-time performance. To address these challenges, this paper proposes LSD-Det, a lightweight SAR ship detection model improved from YOLOv8n. First, the YOLOv8n architecture is streamlined to reduce redundancy, enhancing suitability for SAR detection. Then, a Grouped Split Enhanced Channel Attention (GSECA) module is introduced in the backbone, combining average and max pooling with channel shuffle to improve recognition of small targets and suppress background noise. Additionally, a Global-Enhanced Dilated Wavelet Transform (GEDWT) module is embedded in the neck’s C2f structure to enhance multi-scale feature representation with minimal computational overhead. Furthermore, the original CIoU loss is replaced with PIoUv2, which accelerates convergence and improves bounding box regression accuracy. Experiments conducted on two publicly available SAR ship detection datasets, SSDD and HRSID, demonstrate that LSD-Det significantly reduces computational cost while maintaining high detection performance. Specifically, parameters are reduced by 65.7%, GFLOPs by 20.7%, and inference speed is notably improved. LSD-Det also improves mAP on SSDD by 1.2% (mAP@0.5) and 2.2% (mAP@0.5:0.95), with corresponding gains of 0.8% and 1.3% on HRSID. The experimental outcomes demonstrate that the proposed approach maintains a desirable trade-off between precision and computational efficiency, making it suitable for real-time SAR ship detection under limited-resource conditions.
- 28 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 133483-133496
Open AccessLSD-Det: A Lightweight Detector for Small Ship Targets in SAR Images
Synthetic Aperture Radar (SAR) plays a vital role in ship safety monitoring and marine environmental protection. However, ship targets in SAR images are often small, have blurred edges, and suffer from strong background interference, making it difficult for traditional detection algorithms to balance accuracy and real-time performance. To address these challenges, this paper proposes LSD-Det, a lightweight SAR ship detection model improved from YOLOv8n. First, the YOLOv8n architecture is streamlined to reduce redundancy, enhancing suitability for SAR detection. Then, a Grouped Split Enhanced Channel Attention (GSECA) module is introduced in the backbone, combining average and max pooling with channel shuffle to improve recognition of small targets and suppress background noise. Additionally, a Global-Enhanced Dilated Wavelet Transform (GEDWT) module is embedded in the neck’s C2f structure to enhance multi-scale feature representation with minimal computational overhead. Furthermore, the original CIoU loss is replaced with PIoUv2, which accelerates convergence and improves bounding box regression accuracy. Experiments conducted on two publicly available SAR ship detection datasets, SSDD and HRSID, demonstrate that LSD-Det significantly reduces computational cost while maintaining high detection performance. Specifically, parameters are reduced by 65.7%, GFLOPs by 20.7%, and inference speed is notably improved. LSD-Det also improves mAP on SSDD by 1.2% (mAP@0.5) and 2.2% (mAP@0.5:0.95), with corresponding gains of 0.8% and 1.3% on HRSID. The experimental outcomes demonstrate that the proposed approach maintains a desirable trade-off between precision and computational efficiency, making it suitable for real-time SAR ship detection under limited-resource conditions.
- Open Access
- 9 July 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (14) , 7707
Semantic Attention Guided Hierarchical Decision Network for Age Estimation
Image-based human age estimation has been a hot topic in the field of face attribute recognition. As the greatest dilemma of age estimation is the similar appearance of adjacent age labels, this paper presents a two-stage age estimation model based on the idea of coarse-to-fine decision strategy and fine-grained feature learning mechanism. In the first stage, Swin Transformer is employed to perform global age estimation for each image, and then an adaptive age range is obtained by local adjusted the global age estimation result. This stage gives a relative age trend of the input image. In the second stage, a local regressor is performed on the generated age range based on a sub-network combining semantic attention mechanism and hierarchical fine-grained feature pooling. The proposed semantic attention provides a guidance of age-sensitive regions for distinguishing fine-grained age-relevant feature in the second stage. The formulation of the proposed two-stage framework is intuitive and end-to-end trainable, and experiments on three popular age estimation benchmarks achieve the state-of-the-art results.
- 9 July 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (14) , 7707
Open AccessSemantic Attention Guided Hierarchical Decision Network for Age Estimation
Image-based human age estimation has been a hot topic in the field of face attribute recognition. As the greatest dilemma of age estimation is the similar appearance of adjacent age labels, this paper presents a two-stage age estimation model based on the idea of coarse-to-fine decision strategy and fine-grained feature learning mechanism. In the first stage, Swin Transformer is employed to perform global age estimation for each image, and then an adaptive age range is obtained by local adjusted the global age estimation result. This stage gives a relative age trend of the input image. In the second stage, a local regressor is performed on the generated age range based on a sub-network combining semantic attention mechanism and hierarchical fine-grained feature pooling. The proposed semantic attention provides a guidance of age-sensitive regions for distinguishing fine-grained age-relevant feature in the second stage. The formulation of the proposed two-stage framework is intuitive and end-to-end trainable, and experiments on three popular age estimation benchmarks achieve the state-of-the-art results.
- Open Access
- 24 March 2025
- Published by MDPI AG in Mathematics
- Vol. 13 (7) , 1054
DIA-TSK: A Dynamic Incremental Adaptive Takagi–Sugeno–Kang Fuzzy Classifier
In order to continuously adapt to dynamic data distributions, existing incremental and online learning methods adopt bagging or boosting structures, in which some sub-classifiers are abandoned when the data distribution varies significantly in the learning process. As such, these ensemble classifiers may fail to reach the global optimum. Furthermore, the training of static sub-classifiers, which are dropped when concept drift emerges, leads to unnecessary computational costs. To solve these issues, this study proposes a novel training method consisting of a single dynamic classifier—named the dynamic incremental adaptive Takagi–Sugeno–Kang fuzzy classifier (DIA-TSK)—which leverages the superior non-linear modeling capabilities and interpretability of the TSK fuzzy system. DIA-TSK utilizes a multi-dimensional incremental learning strategy that is capable of dynamically learning from new data in real time while maintaining global optimal solutions across various online application scenarios. DIA-TSK incorporates two distinct learning paradigms: online learning (O-DIA-TSK) and batch incremental learning (B-DIA-TSK). These modules can work separately or collaborate synergistically to achieve rapid, precise and resource-efficient incremental learning. With the implementation of O-DIA-TSK, we significantly reduce the computational complexity in incremental processes, effectively addressing real-time learning requirements for high-frequency dynamic data streams. Moreover, the novel incremental update mechanism of O-DIA-TSK dynamically adjusts its parameters to ensure progressive optimization, enhancing both real-time performance and learning accuracy. For large-scale data sets, DIA-TSK evolves into B-DIA-TSK, which implements batch updates for multiple samples based on the Woodbury matrix identity. This extension substantially improves computational efficiency and robustness during incremental learning, making it particularly suitable for high-dimensional and complex data sets. Extensive comparative experiments demonstrate that the DIA-TSK approaches significantly outperform existing incremental learning methods across multiple dynamic data sets, exhibiting notable advantages in terms of computational efficiency, classification accuracy and memory management. In the experimental comparison, O-DIA-TSK and B-DIA-TSK reach significant superiority in classification performance with respect to comparative methods, with up to 33.3% and 55.8% reductions in training time, respectively, demonstrating the advantage of DIA-TSK in classification tasks using dynamic data.
- 24 March 2025
- Published by MDPI AG in Mathematics
- Vol. 13 (7) , 1054
Open AccessDIA-TSK: A Dynamic Incremental Adaptive Takagi–Sugeno–Kang Fuzzy Classifier
In order to continuously adapt to dynamic data distributions, existing incremental and online learning methods adopt bagging or boosting structures, in which some sub-classifiers are abandoned when the data distribution varies significantly in the learning process. As such, these ensemble classifiers may fail to reach the global optimum. Furthermore, the training of static sub-classifiers, which are dropped when concept drift emerges, leads to unnecessary computational costs. To solve these issues, this study proposes a novel training method consisting of a single dynamic classifier—named the dynamic incremental adaptive Takagi–Sugeno–Kang fuzzy classifier (DIA-TSK)—which leverages the superior non-linear modeling capabilities and interpretability of the TSK fuzzy system. DIA-TSK utilizes a multi-dimensional incremental learning strategy that is capable of dynamically learning from new data in real time while maintaining global optimal solutions across various online application scenarios. DIA-TSK incorporates two distinct learning paradigms: online learning (O-DIA-TSK) and batch incremental learning (B-DIA-TSK). These modules can work separately or collaborate synergistically to achieve rapid, precise and resource-efficient incremental learning. With the implementation of O-DIA-TSK, we significantly reduce the computational complexity in incremental processes, effectively addressing real-time learning requirements for high-frequency dynamic data streams. Moreover, the novel incremental update mechanism of O-DIA-TSK dynamically adjusts its parameters to ensure progressive optimization, enhancing both real-time performance and learning accuracy. For large-scale data sets, DIA-TSK evolves into B-DIA-TSK, which implements batch updates for multiple samples based on the Woodbury matrix identity. This extension substantially improves computational efficiency and robustness during incremental learning, making it particularly suitable for high-dimensional and complex data sets. Extensive comparative experiments demonstrate that the DIA-TSK approaches significantly outperform existing incremental learning methods across multiple dynamic data sets, exhibiting notable advantages in terms of computational efficiency, classification accuracy and memory management. In the experimental comparison, O-DIA-TSK and B-DIA-TSK reach significant superiority in classification performance with respect to comparative methods, with up to 33.3% and 55.8% reductions in training time, respectively, demonstrating the advantage of DIA-TSK in classification tasks using dynamic data.
- Open Access
Multi-Scale Feature Enhancement Method for Underwater Object Detection
With deep-learning-based object detection methods reaching industrial-level performance, underwater object detection has emerged as a significant application. However, it is often challenged by dense small instances and image blurring due to the water medium. In this paper, a Multi-Scale Feature Enhancement(MSFE) method is presented to address the challenges triggered by water bodies. In brief, MSFE attempts to achieve dual multi-scale information integration through the internal structural design of the basic C2F module in the Backbone network and the external global design of the feature pyramid network (FPN). For the internal multi-scale implementation, a LABNK module is constructed to address the vanishing or weakening phenomenon of fine-grained features during feature extraction. Specifically, it adopts a symmetrical structure to collaboratively capture two types of local receptive field information. Furthermore, to enhance the information integration ability between inter-layer features in FPN, a shallow feature branch is injected to supplement detailed features for the subsequent integration of multi-scale features. This operation is mainly supported by the fact that large-sized features from the shallow layer usually carry rich, fine-grained information. Taking the typical YOLOv8n as the benchmark model, extensive experimental comparisons on public underwater datasets (DUO and RUOD) demonstrated the effectiveness of the presented MSFE method. For example, taking the rigorous mAP (50:95) as an evaluation metric, it can achieve an accuracy improvement of about 2.8%.
Open AccessMulti-Scale Feature Enhancement Method for Underwater Object Detection
With deep-learning-based object detection methods reaching industrial-level performance, underwater object detection has emerged as a significant application. However, it is often challenged by dense small instances and image blurring due to the water medium. In this paper, a Multi-Scale Feature Enhancement(MSFE) method is presented to address the challenges triggered by water bodies. In brief, MSFE attempts to achieve dual multi-scale information integration through the internal structural design of the basic C2F module in the Backbone network and the external global design of the feature pyramid network (FPN). For the internal multi-scale implementation, a LABNK module is constructed to address the vanishing or weakening phenomenon of fine-grained features during feature extraction. Specifically, it adopts a symmetrical structure to collaboratively capture two types of local receptive field information. Furthermore, to enhance the information integration ability between inter-layer features in FPN, a shallow feature branch is injected to supplement detailed features for the subsequent integration of multi-scale features. This operation is mainly supported by the fact that large-sized features from the shallow layer usually carry rich, fine-grained information. Taking the typical YOLOv8n as the benchmark model, extensive experimental comparisons on public underwater datasets (DUO and RUOD) demonstrated the effectiveness of the presented MSFE method. For example, taking the rigorous mAP (50:95) as an evaluation metric, it can achieve an accuracy improvement of about 2.8%.