In the field of modern warfare and defense technology, accurate detection of tank targets is crucial for military decision-making and tactical deployment. However, the traditional target detection algorithm has problems such as insufficient detection accuracy and false detection. In order to solve these problems, a target detection algorithm based on improved yolov5 s is proposed to realize the recognition of tank targets. The algorithm designs a network module that can extract multi-scale features, and optimizes the C3 module in yolov5s to enhance the feature extraction ability of the algorithm. An improved feature extraction module based on spatial pyramid pooling is proposed, which captures feature information of different scales by adding pooling kernels of different scales. The iRMB ( Inverted Residual Module with Bottleneck ) module is introduced and improved, and the attention module is replaced by LSKA ( Large Kernel Attention ) to improve the ability of the network to identify key information. The test results on the tank dataset show that the improved algorithm improves the average accuracy by 3.2 %, the recall rate by 2.1 %, the mAP50 by 1.1 %, and the mAP50-95 by 2.5 %. These results fully prove that the improved yolov5 s model has significant performance advantages in tank target detection.