Research On Small Object Detection Algorithm Based On YOLOv5s

With the gradual rise of small target object detection,related technological achievements have significant application value in fields such as military defense,disaster detection,and smart cities.However,small target objects themselves have problems such as large scale differences and low resolution.In the process of collecting small target data,there are also problems such as severe object occlusion and overlap,background element interference,and blurred collected images.To effectively solve the above problems,this article proposes a new model YOLOv5s ours based on the YOLOv5s model,which mainly includes the following three innovative works:(1)Due to the large amount of computation and parameters of the current small target object detection model,this paper introduces the lightweight backbone network Ghost Net,and proposes a lightweight module Ghost ours Module,which greatly reduces the amount of parameters and computation of the benchmark model,and the improved model has a faster training speed.(2)In response to the problem of difficult feature information acquisition in small object detection tasks,the BiFPN(Bidirectional Feature Pyramid Network)bidirectional feature pyramid structure is first adopted at the neck of the model in this paper,followed by the addition of CBAM(Convolutional Block Attention Module)and STR(Swin Transformer)modules.The ablation study shows that this improvement significantly improves the feature extraction ability and detection accuracy of the model.(3)In response to the issues of missed and false detections in small object detection tasks,this model has added an additional detection head on top of the original one,replacing three-scale detection with four-scale detection.The experiment shows that this improvement effectively improves the model’s ability to detect multi-scale targets.