Algorithm Research And System Implementation Of UAV Patrol Image Recognition

With the rapid development of China’s economy,the requirements of power system for power quality and power supply stability are gradually improved.Traditional manual inspection has been unable to meet the stability and timeliness requirements of power system.In recent years,the research direction and production practice direction is to combine UAV technology with aviation remote sensing technology and artificial intelligence image recognition technology to carry out accurate fault detection for overhead transmission lines and transmission towers.In this paper,the most critical artificial intelligence image recognition technology and its system application in UAV patrol inspection are studied.The main work is summarized as follows:First of all,according to a large number of UAV inspection images collected in the field,according to the fault characteristics,it classifies them and establishes the algorithm training data set,which lays the foundation for the subsequent optimization training and performance test of the model algorithm.Then,the single-stage detection algorithm of transmission line inspection image based on self attention mechanism is studied.At the same time,in order to overcome the shortcomings of limited local receptive field of convolutional network and avoid the problem of too large parameters of full connection layer,this paper introduces self attention mechanism in the model,and establishes an improved Yolo V3 algorithm:clustering anchor anchor types with K-means,and adding self attention mechanism in the extraction module to further improve the detection effect of the model.Experiments show that the improved Yolo V3 model not only achieves the performance of two-stage model in accuracy,but also is far better than the two-stage model in detection speed.Finally,based on Django framework,an image management and intelligent recognition system for power inspection is designed and developed.Through the analysis of system requirements,all kinds of objects are modeled and database models are designed,and then the detailed design and implementation of all functional modules of the system are completed,and all functions and performance of the system are tested.The system realizes the object-oriented management of inspection tasks and inspection images,realizes the intellectualization of image interpretation,visualization of interpretation results,and supports the statistical analysis of interpretation results and automatic generation of reports.