Small Target Detection Methods in Complex Scenes

Small target detection is of great significance in military and civilian fields, so the problems of small target aggregation and scale diversity under scene complexity, this paper proposes a small target detection method under complex scenes for small target aggregation and scale diversity under scene complexity. Firstly, we design a multiscale grouping efficient convolution module, Multi-ScaleGroupEfficientConv, to replace part of the C2f module, so as to extract multiscale features efficiently; secondly, we construct a feature pyramid network for small targets and construct a CSP-OmniKernel feature fusion module based on the idea of CSP and OmniKernel to retain the information of small targets to a greater extent. The intent is to retain the information of small targets and improve the target detection ability. Finally, on the VisDrone2019 dataset, comparison experiments are conducted with mainstream algorithms, and the experimental results show that the method in this paper improves 1.2%, 2.1%, 2.6%, and 1.3% in the detection rate, recall rate, mAP<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">50</inf>, and mAP<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">50:95</inf>, respectively, in comparison with the original yolov8n, and the model size and the number of parameters are only 6.1MB and 3.02M, which are better than other mainstream methods.