An advanced YOLOv3 method for small object detection

Small object detection is a very challenging task in the field of object detection because it is easily affected by large object occlusion and small object itself has relatively little feature information. Aiming at the problem that the YOLOv3 network does not consider the context semantic relationship of small object detection, the detection accuracy of small objects is not high. In this paper, we propose a small object detection network combining multi-level fusion and feature augmentation. First, the feature enhancement module is introduced into the deep layer of the backbone extraction network to enhance the feature information of small objects in the feature map. Second, a multi-level feature fusion module is proposed to better capture the contextual semantic relationship of small objects. In addition, the strategy of combining Soft-NMS and CIOU is used to solve the problem of missed detection of occluded small objects. At last, The ablation experiment of the MS COCO2017 object detection task proves the effectiveness of several modules introduced in this paper for small object detection. The experimental results on the MS COCO2017, VOC2007, and VOC2012 datasets show that the AP of this method is 16.5%, 8.71%, and 9.68% higher than that of YOLOv3, respectively. All experiments show that the method proposed in this paper has better detection performance for small object detection.