In order to solve the problems of long monitoring cycle and low degree of intelligence in traditional disc suspension insulator deterioration monitoring, this paper uses the improved YOLOX neural network for insulator and its breakage defect target detection. The network structure is improved, the loss function is modified, and attention mechanism is added to seek the optimal weight of the neural network running speed and prediction accuracy. The training effect of the neural network is improved by expanding the image dataset. Considering the characteristics of low/zero defects of insulators, such as strong concealment and high recognition difficulty, a low/zero value detection method for ceramic insulators based on infrared image color information is adopted. A color feature selection method based on Adjusted Mutual Information(AMI) is proposed to select representative color feature components and conduct threshold analysis, further improving the defect prediction accuracy. The experiments show that the proposed method can monitor the deterioration state of disc suspension insulators throughout the process, with high prediction accuracy and good practicality.