The invention discloses a
traffic accident severity detection method based on
wavelet frequency division enhancement and multistage feedback, a
traffic accident severity detection model based on YOLOv11 is constructed, and a C3k2-DAWTConv module is designed to replace a C3k2 module in an original model structure, so that the sensitivity of the model to detail features is effectively improved while the parameter quantity and the calculation quantity are reduced, and the detection accuracy of the
traffic accident severity is improved. According to the method, significant lightweight and performance balance are realized, the multi-scale
receptive field attention
convolution MSRFAConv is designed in the Neck network, and network parameters are improved while calculation cost and parameters are almost not increased. An original SPPF module is replaced with the FocalModulation technology, the method is suitable for
processing small objects which are difficult to detect or objects in a complex background, through experimental
verification, the detection precision of the method is superior to that of an original conventional target detection model, network parameters and the calculated amount are reduced, and the robustness of the model is improved. The method detects the severity of the traffic accident with low threshold and high precision, and has wide application potential.