Improved garment detection algorithm based on YOLOv5
Fenghua Liu, Hongsheng Zhao, Weiguang Liu · 2022 5th World Conference on Mechanical Engineering and Intelligent Manufacturing (WCMEIM) · 2022
An improved target detection algorithm based on YOLOv5 is proposed to address the problems of high similarity of existing clothing data, complex backgrounds, and low detection accuracy due to the difficult extraction of inter-class features. Based on the characteristics of the DeepFashion2 dataset, a new feature extraction layer with 64-times downsampling is designed, and a four-feature layer is used to detect larger clothing images; a lightweight GhostNet network is used instead of the backbone extraction network to reduce the network parameters and enhance the feature representation; Position information is incorporated into feature extraction to enhance the sensitivity of the model to position information;The experimental results show that compared with the original YOLOv5 network, the improved network models map50 and map50:95 are improved by 3.36% and 4. 7%, respectively, which can meet the practical application requirements of garment detection.