Small Object Detection of Human Face Melasma Based on Improved YOLOv5s

Aiming at the problem that YOLOv5s is used for human face melasma detection because it ignores the detail information leading to the missed detection of small objects of the melasma, this paper presents a small object detection method utilizing an optimized version of YOLOv5s. The YOLOv5s-CCB deep learning network is constructed, and the CA attention is integrated into the C3 module of the backbone network to improve the network's capability in capturing the precise location information of small objects; the CARAFE up-sampling module is incorporated through the neck to enhance the model's ability to extract global features; and the BiFPN bi-directional feature pyramid network is introduced during the feature fusion stage, the goal is to effectively combine shallow and deep features. Experimental tests demonstrate that the mAP_0.5 and R of this method are improved by 3.1 % and 3% respectively, the performance surpasses that of the original YOLOv5s detection method.