Small objects are easy to be lost and misjudged in the detection task because of their relatively low resolution in the image. Aiming at the problem that the detection accuracy of small-scale targets in the current target detection algorithm is much lower than that of other sizes, the feature enhancement of small-scale targets is integrated into the feature pyramid structure to avoid the lack of small-scale feature information. The feature enhancement ability of multi-scale feature fusion is used to enrich the feature information of small-scale target feature layer, so as to improve the accuracy of small-scale target detection. The improved feature pyramid structure is applied to YOLOv3 network. The experimental comparative study shows that the detection accuracy of small-scale targets can reach 0.179, which is 22.6% higher than the original network.