In order to solve the problem of poor performance of deep neural network model in small target detection, the Focus structure is improved, and a plug-and-play lightweight structure Focus + is proposed. Relevant images are collected and a military small target dataset containing 5 types of targets is established. Focus + is inserted into the commonly used target detection model, and several groups of comparative experiments are carried out using different scale inputs. The experimental results show that the average accuracy of model detection is improved by 0. 8% after the introduction of Focus + module, which indicates that it can extract the high-resolution features of shallow network while occupying less computing power, and effectively improve the detection accuracy of small targets.