In the field of small target detection, many algorithms improve the accuracy at the cost of increasing the complexity of the model, but it brings a large computational burden and equipment requirements. Aiming at the contradiction between complexity and detection accuracy in the model, an improved resampling strategy algorithm on image pyramid is proposed. The algorithm only needs to calculate a small amount of sample data and introduce a lightweight attention mechanism module with a few parameters. In the experiment, training and testing were carried out on the COCO dataset. The resampling strategy mAP value was 40.6%, and the improved value was 42.1% with the introduction of the attention module, and the weight file size only increased by 2% with the introduction of the attention module. The experimental results show that the improvement of the sample resampling algorithm can improve the detection accuracy while reducing the computational burden, which verifies the effectiveness of the lightweight attention module.