Abstract: In order to realize the intelligent evaluation of day-blind UV imaging of suspended insulator discharge based on edge detection, this paper established a test platform of suspended insulator pollution discharge and obtained the UV parameters such as relative spot area. Based on this, we put forward the criterion of discharge severity, and established the UV image sample database of different discharge degrees. The improved SSD model was used to train the UV TV frequency stream, which realized the reduction of model parameters and the improvement of training convergence speed. The results show that, when the improved SSD model is used to train the discharge image, the model file size is 4.5mb and has good embedded detection performance. In the process of model training, we designed the self-updating learning rate mechanism to increase the convergence speed of the model by about 3.5 times and reduce the convergence value by 1/2. The intelligent assessment of insulator discharge severity can be divided into four grades: “slow”, “moderate”, “heavy” and “worse”. The results of this paper provide a new idea for the real-time assessment of power equipment discharge severity based on UV imaging on the edge side.