In this paper, the latest YOLOv3 deep network model in the field of artificial intelligence is used to study the accuracy of insulator detection and definition in aerial photography images,proposing a decomposition and aggregation algorithm based on YOLOv3 .In order to solve the problems of error detection and leakage detection of targets in insulator testing,the actual target is decomposed into several continuous and intersecting variable parts, which are detected and identified.In guarantee under the premise of target detection accuracy and speed, using the feature information and the meaning of the intersection zone between parts, parts of belonging to the original target aggregate and redefined, to detect the target definition is more accurate.The original model algorithm cannot be defined accurately because there are too many variables in the group target such as crowd,this paper puts forward the improved method can be according to the necessary components to identify detection, appear as separate sub targets at the same time find out the whole, it belongs with multistage label on the deeper sense of the character description.Take COCO data set as an example to compare the detection effect before and after the improvement of the algorithm. The experimental results show that this method can improve the accuracy of target detection and solve the problems of missed detection and wrong detection.