In order to effectively solve the problem that small objects are difficult to recall and easy to miss detection, an improved YOLOv3 algorithm based on feature fusion and feature enhancement was proposed. To enhance the generalization performance of the model, the Mosaic and Mixup methods were combined for data enhancement during training. Firstly, in order to improve the low recall rate of small object detection, the original feature fusion network was extended to the shallower layer, and the bottom-up feature pyramid was added, so that the details and positioning information of the shallow feature layer could be transmitted to the deep layer. Secondly, a feature enhancement module was proposed to enlarge the receptive field, so that the shallow feature layer could obtain the rich deep semantic information and optimize the expression ability of the feature layer. Finally, the GIoU was used as a regression loss function to reduce the missing rate and achieve more accurate regression. Simulation experiments on Pascal VOC2007 and VOC2012 show that the improved algorithm can improve the mean average precision (mAP) value by 4.4% on the premise of ensuring the detection speed. Experimental results fully prove that the proposed algorithm can effectively improve the performance of small object detection.