Rapid iteration method of intelligent catenary defect detection model

The invention belongs to the technical field of computer models, and particularly relates to a rapid iteration method of an intelligent catenary defect detection model. The method comprises the following steps: step 1, inputting a plurality of catenary images for training, and generating a dynamic convolution kernel based on a connection attention head; step 2, performing defect type grouping on all the catenary images for training, and taking the defect type grouping as a dynamic convolution kernel of the defect type; 3, establishing an intelligent contact network defect detection model, and adjusting the weight of the dynamic convolution kernel of the defect type by using a self-adaptive gradient cutting method; and 4, after the rapid iteration is completed, extracting the dynamic convolution features by using the updated dynamic convolution kernel. The feature extraction capability, the training efficiency and the detection precision of the model are remarkably improved. 本发明属于计算机模型技术领域，具体涉及一种接触网缺陷智能检测模型的快速