With the consumption of fossil fuels，environmental pollution is serious，the permeability of clean and environmental protection renewable energy power generation continues to increase． And with the expansion of wind farm construction in cold areas，it will lead to blade icing，which will seriously lead to blade breakage． Therefore，it is necessary to study the blade icing recognition． Based on the analysis of the existing blade icing detection methods，the edge pixels of the image are recognized by combining the improved Canny algorithm with the salient recognition algorithm，and the pixel gradient of each point is calculated． The optimal threshold is determined by using gray value segmentation method，and after the significant recognition is carried out，the recognition of icing image of the blade is studied． The research shows that the image recognition of blade icing has a high efficiency and reliability，which can help the blade deicing and ensure the safe and reliable operation of wind turbines．