Agricultural insect pest image detection method based on SFP-YoloX

The invention provides an agricultural insect pest image detection method based on SFP-YoloX, and the method comprises the steps: replacing a part of a backbone network darknet53 feature extraction layer of an original model YoloX with a Swinin-Transform structure, and enabling a module to better capture global features; according to the research, an existing latest method is combined to improve an original spatial pyramid method, a channel attention and spatial attention mechanism module GAM is added, and fusion of multiple receptive fields is improved; 1 * 1 convolution before up-sampling is canceled in an original FPN feature pyramid structure; a C3 module of an original model is replaced by a C2f module, so that the learning ability of the model for deep features is further improved; a prediction head for micro object detection is added in a check in an original model, and the four-head structure can relieve negative effects caused by severe target scale changes, so that a better training result of the im