부산대학교 도서관

In this paper, a lightweight pedestrian detection algorithm named GGSnet algorithm based on improved YOLOv5s is proposed to improve the detection speed and accuracy of pedestrian recognition, and a comprehensive experimental evaluation is carried out on pedestrian recognition tasks. Ghost module is introduced into YOLOv5s to make the model highly computative while maintaining lightweight. GSConv module replaces traditional convolution operation, which can effectively reduce the complexity of the model, improve the computational efficiency, and maintain the accuracy to a certain extent. At the same time, the SIMAM module is introduced to improve the feature extraction capability of the image and further improve the accuracy of pedestrian recognition. Experimental results show that GGSnet achieves high Precision and mAP scores of 0.915 and 0.932, respectively, demonstrating the advantages of the algorithm in accuracy. In addition, GGSnet has a reference count of 3366,526 and a model size of 7.1MB, 48% and 49.3% of YOLOv5s, respectively, which helps achieve efficient pedestrian recognition in environments with limited storage and computing resources. This verifies that GGSnet has potential in the field of pedestrian detection and provides valuable exploration in practical applications. Therefore, the pedestrian recognition algorithm based on GGSnet proposed in this paper can be used as a promising solution and provide new research ideas and technical support for other fields.