Methods of image recognition of overhead power line insulators and ice types based on deep weakly‐supervised and transfer learning

Insulator icing endangers the safe operation of overhead power lines. The current measures for image monitoring lines rely on manual observation. There is an urgent need intelligent analysis methods with strong generalisation, accuracy, and efficiency to improve grid decision-making. Based data from China Southern Power Grid, this paper first proposes rules insulator cleaning, classification, annotation, establishing a dataset images during 2014–2018. Then, based Yolo (You only look once) v5, weakly supervised phased transfer learning method proposed recognise insulators ice types such as snow, rime, mixed glaze, normal. This reduces interference background, uneven illumination, camera occlusion through multidimensional feature fusion. precision, recall, mean average precision recognition can reach 86.6%, 91.3%, 90.1%, respectively, speed fast 8 ms/image. Using pseudo-labelling enables intellectualisation which significantly changes inefficiency labelling.