This paper examines the atmospheric corrosion characteristics of power metal frame equipment in Chongqing Power Grid. The "exposure" experiment of Q235 steel was carried out in the substation, and the on-site corrosion data of Q235 steel with different corrosion degrees were obtained. For metal corrosion morphology at different exposure time, different data were measured, including the color, hue and saturation related features which were based on HSI and RGB image models: MR, MG, MB, σR, σG, σB, MH, MS, σH, σS; statistical characteristics: corrosion energy E, corrosion entropy S;binary characteristic: corrosion rate k; fractal feature: fractal dimension FD. A total of 14 features were selected as the evaluation system of corrosion feature vector. We use the digital image processing technology and the BP neural network algorithm to qualitatively evaluate the corrosion state of equipment metal. By testing the on-site samples of the other 3 substations, the corrosion state values of the samples were 1.0293, 0.9967 and 0.8508 respectively, which were basically consistent with the actual corrosion degree. The system had a good evaluation effect.