Most Downloaded Articles

Published in last 1 year | In last 2 years| In last 3 years| All| Most Downloaded in Recent Month | Most Downloaded in Recent Year|

Most Downloaded in Recent Year
Please wait a minute...
Survey of deep learning based EEG data analysis technology
Bo ZHONG,Pengfei WANG,Yiqiao WANG,Xiaoling WANG
Journal of ZheJiang University (Engineering Science)    2024, 58 (5): 879-890.   DOI: 10.3785/j.issn.1008-973X.2024.05.001
Abstract   HTML PDF (690KB) ( 2360 )  

A thorough analysis and cross-comparison of recent relevant works was provided, outlining a closed-loop process for EEG data analysis based on deep learning. EEG data were introduced, and the application of deep learning in three key stages: preprocessing, feature extraction, and model generalization was unfolded. The research ideas and solutions provided by deep learning algorithms in the respective stages were delineated, including the challenges and issues encountered at each stage. The main contributions and limitations of different algorithms were comprehensively summarized. The challenges faced and future directions of deep learning technology in handling EEG data at each stage were discussed.

Table and Figures | Reference | Related Articles | Metrics
Research overview on touchdown detection methods for footed robots
Xiaoyong JIANG,Kaijian YING,Qiwei WU,Xuan WEI
Journal of ZheJiang University (Engineering Science)    2024, 58 (2): 334-348.   DOI: 10.3785/j.issn.1008-973X.2024.02.012
Abstract   HTML PDF (1751KB) ( 981 )  

The effects of leg structure design, foot-end design and sensor design on touchdown detection were comprehensively discussed by analyzing the existing legged robot touchdown detection methods. The touchdown method for direct detection of external sensors, the touchdown detection method based on kinematics and dynamics, and the touchdown detection method based on learning were summarized. Touchdown detection methods were summarized in three special scenarios: slippery ground, soft ground, and non-foot-end contact. The application scenarios of touchdown detection technology were analyzed, including the three application scenarios of motion control requirements, navigation applications, and terrain and geological sensing. The development trends were pointed out, which related to the four major touchdown detection methods of hardware improvement and integration, multi-mode touchdown detection, multi-sensor fusion touchdown detection, and intelligent touchdown detection. The specific relationships between various touchdown detection algorithms were summarized, which provided guidance for the development of follow-up technology for touchdown detection and specific applications of touchdown detection.

Table and Figures | Reference | Related Articles | Metrics
Multimodal sentiment analysis model based on multi-task learning and stacked cross-modal Transformer
Qiao-hong CHEN,Jia-jin SUN,Yang-bo LOU,Zhi-jian FANG
Journal of ZheJiang University (Engineering Science)    2023, 57 (12): 2421-2429.   DOI: 10.3785/j.issn.1008-973X.2023.12.009
Abstract   HTML PDF (1171KB) ( 665 )  

A new multimodal sentiment analysis model (MTSA) was proposed on the basis of cross-modal Transformer, aiming at the difficult retention of the modal feature heterogeneity for single-modal feature extraction and feature redundancy for cross-modal feature fusion. Long short-term memory (LSTM) and multi-task learning framework were used to extract single-modal contextual semantic information, the noise was removed and the modal feature heterogeneity was preserved by adding up auxiliary modal task losses. Multi-tasking gating mechanism was used to adjust cross-modal feature fusion. Text, audio and visual modal features were fused in a stacked cross-modal Transformer structure to improve fusion depth and avoid feature redundancy. MTSA was evaluated in the MOSEI and SIMS data sets, results show that compared with other advanced models, MTSA has better overall performance, the accuracy of binary classification reached 83.51% and 84.18% respectively.

Table and Figures | Reference | Related Articles | Metrics
Multi-behavior aware service recommendation based on hypergraph graph convolution neural network
Jia-wei LU,Duan-ni LI,Ce-ce WANG,Jun XU,Gang XIAO
Journal of ZheJiang University (Engineering Science)    2023, 57 (10): 1977-1986.   DOI: 10.3785/j.issn.1008-973X.2023.10.007
Abstract   HTML PDF (1380KB) ( 654 )  

A multi-behavior aware service recommendation method based on hypergraph graph convolutional neural network (MBSRHGNN) was proposed to resolve the problem of insufficient high-order service feature extraction in existing service recommendation methods. A multi-hypergraph was constructed according to user-service interaction types and service mashups. A dual-channel hypergraph convolutional network was designed based on the spectral decomposition theory with functional and structural properties of multi-hypergraph. Chebyshev polynomial was used to approximate hypergraph convolution kernel to reduce computational complexity. Self-attention mechanism and multi-behavior recommendation methods were combined to measure the importance difference between multi-behavior interactions during the hypergraph convolution process. A hypergraph pooling method named HG-DiffPool was proposed to reduce the feature dimensionality. The probability distribution for recommending different services was learned by integrating service embedding vector and hypergraph signals. Real service data was obtained by the crawler and used to construct datasets with different sparsity for experiments. Experimental results showed that the MBSRHGNN method could adapt to recommendation scenario with highly sparse data, and was superior to the existing baseline methods in accuracy and relevance.

Table and Figures | Reference | Related Articles | Metrics
Survey of embodied agent in context of foundation model
Songyuan LI,Xiangwei ZHU,Xi LI
Journal of ZheJiang University (Engineering Science)    2025, 59 (2): 213-226.   DOI: 10.3785/j.issn.1008-973X.2025.02.001
Abstract   HTML PDF (841KB) ( 347 )  

Foundational models in natural language processing, computer vision and multimodal learning have achieved significant breakthroughs in recent years, showcasing the potential of general artificial intelligence. However, these models still fall short of human or animal intelligence in areas such as causal reasoning and understanding physical commonsense. This is because these models primarily rely on vast amounts of data and computational power, lacking direct interaction with and experiential learning from the real world. Many researchers are beginning to question whether merely scaling up model size is sufficient to address these fundamental issues. This has led the academic community to reevaluate the nature of intelligence, suggesting that intelligence arises not just from enhanced computational capabilities but from interactions with the environment. Embodied intelligence is gaining attention as it emphasizes that intelligent agents learn and adapt through direct interactions with the physical world, exhibiting characteristics closer to biological intelligence. A comprehensive survey of embodied artificial intelligence was provided in the context of foundational models. The underlying technical ideas, benchmarks, and applications of current embodied agents were discussed. A forward-looking analysis of future trends and challenges in embodied AI was offered.

Table and Figures | Reference | Related Articles | Metrics
Surface defect detection algorithm of electronic components based on improved YOLOv5
Yao ZENG,Fa-qin GAO
Journal of ZheJiang University (Engineering Science)    2023, 57 (3): 455-465.   DOI: 10.3785/j.issn.1008-973X.2023.03.003
Abstract   HTML PDF (1697KB) ( 1023 )  

For the poor real-time detection capability of the current object detection model in the production environment of electronic components, GhostNet was used to replace the backbone network of YOLOv5. And for the existence of small objects and objects with large scale changes on the surface defects of electronic components, a coordinate attention module was added to the YOLOv5 backbone network, which enhanced the sensory field while avoiding the consumption of large computational resources. The coordinate information was embedded into the channel attention to improve the object localization of the model. The feature pyramid networks (FPN) structure in the YOLOv5 feature fusion module was replaced with a weighted bi-directional feature pyramid network structure, to enhance the fusion capability of multi-scale weighted features. Experimental results on the self-made defective electronic component dataset showed that the improved GCB-YOLOv5 model achieved an average accuracy of 93% and an average detection time of 33.2 ms, which improved the average accuracy by 15.0% and the average time by 7 ms compared with the original YOLOv5 model. And the improved model can meet the requirements of both accuracy and speed of electronic component surface defect detection.

Table and Figures | Reference | Related Articles | Metrics
Driver fatigue state detection method based on multi-feature fusion
Hao-jie FANG,Hong-zhao DONG,Shao-xuan LIN,Jian-yu LUO,Yong FANG
Journal of ZheJiang University (Engineering Science)    2023, 57 (7): 1287-1296.   DOI: 10.3785/j.issn.1008-973X.2023.07.003
Abstract   HTML PDF (1481KB) ( 895 )  

The improved YOLOv5 object detection algorithm was used to detect the facial region of the driver and a multi-feature fusion fatigue state detection method was established aiming at the problem that existing fatigue state detection method cannot be applied to drivers under the epidemic prevention and control. The image tag data including the situation of wearing a mask and the situation without wearing a mask were established according to the characteristics of bus driving. The detection accuracy of eyes, mouth and face regions was improved by increasing the feature sampling times of YOLOv5 model. The BiFPN network structure was used to retain multi-scale feature information, which makes the prediction network more sensitive to targets of different sizes and improves the detection ability of the overall model. A parameter compensation mechanism was proposed combined with face keypoint algorithm in order to improve the accuracy of blink and yawn frame number. A variety of fatigue parameters were fused and normalized to conduct fatigue classification. The results of the public dataset NTHU and the self-made dataset show that the proposed method can recognize the blink and yawn of drivers both with and without masks, and can accurately judge the fatigue state of drivers.

Table and Figures | Reference | Related Articles | Metrics
Research progress of porous materials with low dielectric constant
WANG Jia-Bang, ZHANG Guo-Quan
J4    2009, 43 (5): 957-961.   DOI: 10.3785/j.issn.1008-973X.2009.05.033
Abstract   PDF (699KB) ( 1941 )  

The porous materials with low dielectric constant are suitable for the applications in integrated circuits. From the aspects of composition and structure, preparation method and dielectric properties, this work introduced the porous low-dielectric-constant materials with different matrix such as inorganic materials, organic materials, inorganic and organic composite separately, whose dielectric constants can be reduced to 1.99, 1.50, 1.99, respectively. The using temperature of the porous low-dielectric-constant materials with organic matrix can reach 450 ℃. The flexural strength of the porous low-dielectric-constant materials with inorganic matrix can reach 136 MPa. The introduction of cave into the materials leads to the decrease of mechanical properties and the increase of dielectric loss. The effort to get a low-dielectric-constant and improve the above properties can broaden the application scope of the porous materials.

Reference | Related Articles | Metrics
Choice of innovation type for China's industrial green transformation under environmental regulation
Haiying LIU,Xianzhe CAI
Journal of ZheJiang University (Engineering Science)    2024, 58 (1): 188-196.   DOI: 10.3785/j.issn.1008-973X.2024.01.020
Abstract   HTML PDF (707KB) ( 584 )  

A super-efficient SBM model including non-desired outputs was used to measure industrial environmental efficiency in 30 Chinese provinces from 2008 to 2020 in order to solve the problem of how industrial enterprises can pick appropriate green technology innovations to accomplish industrial green transformation under the background of strict environmental regulations. The efficiency was used to characterize the level of industrial green transformation. A panel threshold model was used to explore the mechanism of the impact of different green technology innovations on industrial green transformation under different environmental regulation intensities. Results show that China's industrial environmental efficiency fluctuates and rises from 2008 to 2020 as a whole, and the efficiency gap between regions shows a slightly decreasing trend. The environmental impacts of various green technology innovations significantly differ, among which process-oriented green technology innovations emphasizing on processes and products is the key to achieving industrial green transformation. The positive environmental effect of process-oriented green technology innovation increases, while the negative environmental effect of result-oriented green technology innovation decreases as environmental regulations become more stringent.

Table and Figures | Reference | Related Articles | Metrics
Multi-target tracking of vehicles based on optimized DeepSort
Li-sheng JIN,Qiang HUA,Bai-cang GUO,Xian-yi XIE,Fu-gang YAN,Bo-tao WU
Journal of ZheJiang University (Engineering Science)    2021, 55 (6): 1056-1064.   DOI: 10.3785/j.issn.1008.973X.2021.06.005
Abstract   HTML PDF (1014KB) ( 1357 )  

A front multi-vehicle target tracking algorithm optimized by DeepSort was proposed in order to improve the awareness of autonomous vehicles to the surrounding environment. Gaussian YOLO v3 model was adopted as the front-end target detector, and training was based on DarkNet-53 backbone network. Gaussian YOLO v3-Vehicle, a detector specially designed for vehicles was obtained, which improved the vehicle detection accuracy by 3%. The augmented VeRi data set was proposed to conduct the re-recognition pre-training in order to overcome the shortcomings that the traditional pre-training model doesn't target vehicles. A new loss function combining the central loss function and the cross entropy loss function was proposed, which can make the target features extracted by the network become better in-class aggregation and inter-class resolution. Actual road videos in different environments were collected in the test part, and CLEAR MOT evaluation index was used for performance evaluation. Results showed a 1% increase in tracking accuracy and a 4% reduction in identity switching times compared with the benchmark DeepSort YOLO v3.

Table and Figures | Reference | Related Articles | Metrics
Multi-agent pursuit and evasion games based on improved reinforcement learning
Ya-li XUE,Jin-ze YE,Han-yan LI
Journal of ZheJiang University (Engineering Science)    2023, 57 (8): 1479-1486.   DOI: 10.3785/j.issn.1008-973X.2023.08.001
Abstract   HTML PDF (1158KB) ( 1024 )  

A multi-agent reinforcement learning algorithm based on priority experience replay and decomposed reward function was proposed in multi-agent pursuit and evasion games. Firstly, multi-agent twin delayed deep deterministic policygradient algorithm (MATD3) algorithm based on multi-agent deep deterministic policy gradient algorithm (MADDPG) and twin delayed deep deterministic policy gradient algorithm (TD3) was proposed. Secondly, the priority experience replay was proposed to determine the priority of experience and sample the experience with high reward, aiming at the problem that the reward function is almost sparse in the multi-agent pursuit and evasion problem. In addition, a decomposed reward function was designed to divide multi-agent rewards into individual rewards and joint rewards to maximize the global and local rewards. Finally, a simulation experiment was designed based on DEPER-MATD3. Comparison with other algorithms showed that DEPER-MATD3 algorithm solved the over-estimation problem, and the time consumption was improved compared with MATD3 algorithm. In the decomposed reward function environment, the global mean rewards of the pursuers were improved, and the pursuers had a greater probability of chasing the evader.

Table and Figures | Reference | Related Articles | Metrics
Review of underground pipeline monitoring research based on distributed fiber optic sensing
Hai-ying WU,Hong-hu ZHU,Bao ZHU,He QI
Journal of ZheJiang University (Engineering Science)    2019, 53 (6): 1057-1070.   DOI: 10.3785/j.issn.1008-973X.2019.06.005
Abstract   HTML PDF (1057KB) ( 1460 )  

Outline the important role of underground pipelines in national economy and defense construction, as well as the possible serious consequences of pipeline failure. Point out that the real-time monitoring of underground pipelines by using distributed fiber optic sensing (DFOS) technology can guarantee the structural health and safe operation of pipelines. Introduce the pipeline monitoring principle based on DFOS technology, and the research progress of DFOS technology in pipeline leakage monitoring, third party intrusion monitoring, deformation monitoring, corrosion monitoring, geological and natural disaster monitoring and submarine pipeline monitoring. Analyze some existing problems and hot topics in the current research, as well as the future research trend.

Table and Figures | Reference | Related Articles | Metrics
Three-dimensional sector automatic design based on improved NSGA-II algorithm
Yingfei ZHANG,Xiaobing HU,Hang ZHOU,Xuzeng FENG
Journal of ZheJiang University (Engineering Science)    2025, 59 (2): 413-422.   DOI: 10.3785/j.issn.1008-973X.2025.02.019
Abstract   HTML PDF (1634KB) ( 206 )  

An improved non-dominated sorting genetic algorithm II (NSGA-II) was proposed in order to address the challenges of time-consuming manual airspace sectorization and the difficulty in comparing the quality of different sectorization schemes. A three-dimensional multi-objective optimization model for sectorization was established by using a grid-region-sector hierarchy in order to balance controllers’ workload within sectors and reduce workload differences between sectors. A fitness evaluation operator, a probability-adaptive combination crossover operator and a dynamic mutation operator were incorporated in the NSGA-II algorithm in order to enhance the number of feasible solutions, solution diversity and computational efficiency. A simulation was conducted for the automatic 3D sectorization of Xi'an high-altitude airspace. Results showed that the optimized scheme improved workload balance within sectors by 37% and reduced inter-sector workload by 24% compared with the current sectorization configuration. The proposed improved NSGA-II provided a broader range of options for decision-makers with varying preferences compared with traditional weighted multi-objective optimization algorithms.

Table and Figures | Reference | Related Articles | Metrics
Review of Chinese font style transfer research based on deep learning
Ruo-ran CHENG,Xiao-li ZHAO,Hao-jun ZHOU,Han-chen YE
Journal of ZheJiang University (Engineering Science)    2022, 56 (3): 510-519, 530.   DOI: 10.3785/j.issn.1008-973X.2022.03.010
Abstract   HTML PDF (874KB) ( 863 )  

The research works of Chinese font style transfer were classified according to different stages of research development. The traditional methods were briefly reviewed and the deep learning-based methods were combed and analyzed. The commonly used open data sets and evaluation criteria were introduced. The future research trends were expected from four aspects, which were to improve the generation quality, enhance personalized differences, reduce the number of training samples, and learn calligraphy font style.

Table and Figures | Reference | Related Articles | Metrics
Solution approach of Burgers-Fisher equation based on physics-informed neural networks
Jian XU,Hai-long ZHU,Jiang-le ZHU,Chun-zhong LI
Journal of ZheJiang University (Engineering Science)    2023, 57 (11): 2160-2169.   DOI: 10.3785/j.issn.1008-973X.2023.11.003
Abstract   HTML PDF (1371KB) ( 696 )  

Physical information was divided into rule information and numerical information, in order to explore the role of physical information in training neural network when solving differential equations with physics-informed neural network (PINN). The logic of PINN for solving differential equations was explained, as well as the data-driven approach of physical information and neural network interpretability. Synthetic loss function of neural network was designed based on the two types of information, and the training balance degree was established from the aspects of training sampling and training intensity. The experiment of solving the Burgers-Fisher equation by PINN showed that PINN can obtain good solution accuracy and stability. In the training of neural networks for solving the equation, numerical information of the Burgers-Fisher equation can better promote neural network to approximate the equation solution than rule information. The training effect of neural network was improved with the increase of training sampling, training epoch, and the balance between the two types of information. In addition, the solving accuracy of the equation was improved with the increasing of the scale of neural network, but the training time of each epoch was also increased. In a fixed training time, it is not true that the larger scale of the neural network, the better the effect.

Table and Figures | Reference | Related Articles | Metrics
Interface opening strategy of high-speed railway station buildings in response to climate and verification by simulation
Nan WANG,Jin-liu WANG,Cong-hong LIU
Journal of ZheJiang University (Engineering Science)    2023, 57 (6): 1071-1079.   DOI: 10.3785/j.issn.1008-973X.2023.06.002
Abstract   HTML PDF (1898KB) ( 397 )  

A design strategy through appropriately opening the passing space interface in response to climate was proposed as a solution to high energy consumption and formal convergence within the high-speed railway station buildings. Performance simulation tools based on building information modeling (BIM) were used to build a typical model, and the opening time period for different climate zones were determined according to wind and thermal environment simulation analysis. Results show that it is feasible to open up the passing space interface, meeting the requirement of indoor thermal comfort, in the case of a typical summer calculation day in climate zones except in hot summer and warm winter zone (Guangzhou as an example). Meanwhile, during the particular time periods of a year, interface opening is beneficial to energy savings and emission reduction in station buildings, especially in hot summer and cold winter zone (Shanghai as an example) and cold zone (Beijing as an example). The energy-savings reached up to 44.8% and 32.2%, respectively, as well as carbon reduction rates of 36.1% and 21.3%. Hence, the proposed strategy has significant application potential in the green design schemes of high-speed railway station buildings and can explore ideas for regional expression of spatial forms.

Table and Figures | Reference | Related Articles | Metrics
Spatial-temporal multi-graph convolution for traffic flow prediction by integrating knowledge graphs
Jinye LI,Yongqiang LI
Journal of ZheJiang University (Engineering Science)    2024, 58 (7): 1366-1376.   DOI: 10.3785/j.issn.1008-973X.2024.07.006
Abstract   HTML PDF (1616KB) ( 332 )  

A spatial-temporal multi-graph convolution traffic flow prediction model by integrating static and dynamic knowledge graphs was proposed, as current traffic flow prediction methods focus on the spatial-temporal correlation of traffic information and fail to fully take into account the influence of external factors on traffic. An urban traffic knowledge graph and four road network topological graphs with distinct semantics were systematically constructed, drawing upon the road traffic information and the external factors. The urban traffic knowledge graph was inputted into the relational evolution graph convolutional neural network to realize the knowledge embedding. The traffic flow matrix and the knowledge embedding were integrated using the knowledge fusion module. The four road network topology graphs and the traffic flow matrix with fused knowledge were fed into the spatial-temporal multi-graph convolution module to extract spatiotemporal features, and the traffic flow prediction value was outputted through the fully connected layer. The model performance was evaluated on a Hangzhou traffic data set. Compared with the advanced baseline, the performance of the proposed model improved by 5.76%-10.71%. Robustness experiment results show that the proposed model has a strong ability to resist interference.

Table and Figures | Reference | Related Articles | Metrics
Obstacle recognition of unmanned rail electric locomotive in underground coal mine
Tun YANG,Yongcun GUO,Shuang WANG,Xin MA
Journal of ZheJiang University (Engineering Science)    2024, 58 (1): 29-39.   DOI: 10.3785/j.issn.1008-973X.2024.01.004
Abstract   HTML PDF (2463KB) ( 445 )  

The PDM-YOLO model for accurate real-time obstacle detection in unmanned electric locomotives was proposed in order to address the problem of low accuracy of obstacle recognition in existing coal mine underground unmanned electric locomotives due to poor roadway environments. The ordinary convolution in the C3 module of the conventional YOLOv5 model was replaced with partial convolution to construct the C3_P feature extraction module, which effectively reduced the floating-point operations (FLOPs) and computational delay of the model. The improved decoupled head was used to decouple the prediction head of the conventional YOLOv5 model in order to improve the convergence speed of the model and the accuracy of obstacle recognition. The Mosaic data augmentation method was optimized to enrich the feature information of the training images and enhance the generalizability and robustness of the model. The experimental results showed that the mean average precision (mAP) of the PDM-YOLO model reached 96.3% and the average detection speed reached 109.2 frames per second on the self-built dataset. The detection accuracy of the PDM-YOLO model on the PASCAL VOC public dataset is higher than that of the existing mainstream YOLO series models.

Table and Figures | Reference | Related Articles | Metrics
Uncertain behavior sequence prediction method based on intent identification
Fei HE,Cang-hong JIN,Ming-hui WU
Journal of ZheJiang University (Engineering Science)    2022, 56 (2): 254-262.   DOI: 10.3785/j.issn.1008-973X.2022.02.005
Abstract   HTML PDF (1039KB) ( 497 )  

An graph based intent identification embedding (G2IE) method was proposed, in order to solve the problems of behavior uncertainty and data sparsity faced by collaborative recommendation and sequence representation methods in user behavior prediction. In G2IE method, firstly the theory of planned behavior (TPB) is used to mine the controlled behavior patterns in the user behavior sequence, then the transfer intention intensity of the uncertain behavior list between adjacent controlled behaviors is calculated based on information entropy, and finally the behavior relationship is strengthened by integrating the behavior transfer intention to make up for the lack of behavior intention. In G2IE method, the uncertainty of behavior is identified and it is measured with a model, in order to solve the problem of behavior randomness. The problem of data sparsity can be alleviated to some extent by discovering more behavior relationships through the fusion of transfer intention. G2IE method has more accurate and rich expression ability compared with other methods that use behavior direct relation. Experimental results on three public user behavior datasets demonstrate the effectiveness of the proposed method.

Table and Figures | Reference | Related Articles | Metrics
UAV small target detection algorithm based on improved YOLOv5s
Yaolian SONG,Can WANG,Dayan LI,Xinyi LIU
Journal of ZheJiang University (Engineering Science)    2024, 58 (12): 2417-2426.   DOI: 10.3785/j.issn.1008-973X.2024.12.001
Abstract   HTML PDF (708KB) ( 418 )  

An unmanned aerial vehicle (UAV) small target detection algorithm based on YOLOv5, termed FDB-YOLO, was proposed to address the significant issue of misidentification and omissions in traditional target detection algorithms when applied to UAV aerial photography of small targets. Initially, a small target detection layer was added on the basis of YOLOv5, and the feature fusion network was optimized to fully leverage the fine-grained information of small targets in shallow layers, thereby enhancing the network’s perceptual capabilities. Subsequently, a novel loss function, FPIoU, was introduced, which capitalized on the geometric properties of anchor boxes and utilized a four-point positional bias constraint function to optimize the anchor box positioning and accelerate the convergence speed of the loss function. Furthermore, a dynamic target detection head (DyHead) incorporating attention mechanism was employed to enhance the algorithm’s detection capabilities through increased awareness of scale, space, and task. Finally, a bi-level routing attention mechanism (BRA) was integrated into the feature extraction phase, selectively computing relevant areas to filter out irrelevant regions, thereby improving the model’s detection accuracy. Experimental validation conducted on the VisDrone2019 dataset demonstrated that the proposed algorithm outperformed the YOLOv5s baseline in terms of Precision by an increase of 3.7 percentage points, Recall by an increase of 5.1 percentage points, mAP50 by an increase of 5.8 percentage points, and mAP50:95 by an increase of 3.4 percentage points, showcasing superior performance compared to current mainstream algorithms.

Table and Figures | Reference | Related Articles | Metrics