Article Details

Abstract

Aiming at the problems of inadequate spatio-temporal information modeling and poor robustness in complex scenes in existing violence detection methods based on Deep Learning, a violence detection algorithm based on YOLO and hybrid Attention Mechanism is proposed. The algorithm integrates orthogonal channel attention OrthoNets into the C2f module in YOLO backbone network, and uses randomly initialized orthogonal filters to compress feature space information, ensuring feature diversity for subsequent extraction. The channel space mixed attention module CBAM is integrated into the Neck end to enhance the network feature representation capability. The experimental results demonstrate that the algorithm achieves accuracy, recall, and average precision of 91.25%, 95.6%, and 87.5%, respectively, on the RWF-2000 dataset characterized by diverse content and complex environments. It outperforms the most advanced algorithms of the same kind at present in both accuracy metrics and robustness, fully validating its effectiveness.

Keywords

Deep Learning; violence detection; YOLO; Attention Mechanism

References

1. [1] MUMTAZ N，EJAZ N，HABIB S，et al. An Overview of Violence Detection Techniques: Current Challenges and Future Directions [J]. Artificial intelligence review，2023，56（5）：4641-4666.
2. [2] Deer are naturally found in Feng Qin, Yang Huizhong, and others Human behavior recognition based on saliency detection and dense trajectories [J]. Computer Engineering and Applications, 2018, 54 (14): 163-167179
3. [3] FATH U M U，AMIN U，KHAN M，et al.Violence Detection Using Spatiotemporal Features with 3D Convolutional Neural Network [J].Sensors，2019，19（11）：2472.
4. [4] DINESH J S R，FENIL E，GUNASEKARAN M，et al.Real Time Violence Detection Framework for Football Stadium Comprising of Big Data Analysis and Deep Learning Through Bidirectional LSTM [J].Computer Networks，2019，151：191-200.
5. [5] Li Guan, Pang Yulin, Tian Kun Violent video detection based on YOLO and ConvLSTM hybrid neural network [J]. Computer Applications and Software, 2023, 40 (11): 233-240
6. [6] Wang Bin, Xu Honghua, Sun Doucheng, etc Improved small target traffic sign detection based on YOLOv8 algorithm [J]. Computer Applications, 2024, 44 (S2): 274-279
7. [7] SOHAN M，SAI RAM T，RAMI REDDY C V. A Review on Yolov8 and Its Advancements [C]// International Conference on Data Intelligence and Cognitive Informatics.Singapore：Springer，2024：529-545.
8. [8] SALMAN H，PARKS C，SWAN M，et al. Orthonets: Orthogonal Channel Attention Networks [C]// 2023 IEEE International Conference on Big Data (BigData).Sorrento：IEEE，2023：829-837.
9. [9] LIU Q，LIU R，ZHENG B，et al. Infrared Small Target Detection with Scale and Location Sensitivity [C]//Proceedings of the IEEE/CVF 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Seattle：IEEE，2024：17490-17499.
10. [10] Dong Hongzhao, Lin Shaoxuan, She Yini Research progress on YOLO detection technology for traffic targets [J]. Journal of Zhejiang University (Engineering Edition), 2025, 59 (2): 249-260
11. [11] ALDAHOUL N，KARIM H A，DATTA R，et al. Convolutionalneural Network Long Short Term Memory Based IoT Node for Violence Detection [C]//2021 IEEE International Conference onArtificial Intelligence in Engineering and Technology (IICAIET). New York：IEEE，2021：1-6.
12. [12] ULLAH F U M， MUHAMMAD K， HAQ I U， et al. AI-assistededge Vision for Violence Detection in IoT-based Industrial Sur-veillance Networks [J]. IEEE Transactions on Industrial Informatics，2022，18（8）：5359-5370..
13. [13] HUSZÁR V D， ADHIKARLA V K， NEGYESI I，et al. Towardfast and Accurate Violence Detection for Automated Video Surveillance Applications [J].IEEE access，2023，11：18772-18793.
14. [14] Li Jincheng, Yan Ruiao, Dai Xuejing Violent behavior detection based on improved attention mechanism and VGG BiLSTM [J]. Modern Electronic Technology, 2024, 47 (21): 131-138

Funding

Guangxi University for Nationalities Autonomous Region level College Students Innovation and Entrepreneurship Training Program Project (S202310608294); Graduate Education Innovation Project of Guangxi University for Nationalities (gxmzu-chxs2024302)

About the authors

Wu Jiaming (2000-), male, Han ethnicity, from Meizhou, Guangdong, master's student, research direction: image processing and intelligent systems; Lin Ke (2003-), male, Han ethnicity, from Nanning, Guangxi, bachelor's degree, research direction: image processing and intelligent systems; Feng Nan (2002-), female, Manchu, from Chengde, Hebei Province, undergraduate degree, research direction: image processing and intelligent systems; Qiu Sisi (2001-), female, Han ethnicity, from Qinzhou, Guangxi, bachelor's degree, research direction: image processing and intelligent systems; Corresponding author: Wen Yong (1969-), male, Zhuang nationality, from Nanning, Guangxi, associate professor, master's degree, research direction: image processing and intelligent systems.