Suspension porcelain insulator is widely used in power system

and its insulation performance may decrease and thus will seriously threaten the safe and stable operation of power system after a long-term operation in electrical

thermal and stress environments. Therefore

the research on discharge assessment is urgently needed. In this paper

a software and hardware platform for insulator pollution discharge assessment based on UV imaging method is built. The ultraviolet quantization and discharge parameters such as relative UV count

relative spot area

and relative leakage current are obtained. The loss-function-improved YOLOv3 is selected to evaluate the discharge state

and the network parameter optimization based on network super parameters and activation function is realized. The main conclusions are as follows: with the UV quantitative parameters and discharge parameters being defined

four discharge states of "insulator"

"medium"

"bad" and "worse" are classified with a good applicability. After 1 000 iterations

the network error is reduced by 88.5% compared with that obtained from the cross entropy loss function and focal loss function in the train and test period. Based on the study of the characterization parameters such as Class

mA P

and Recall

the adjustment strategy and the selection principle of activation function are confirmed

and the accuracy on the test set is improved by 2.5%. The method and conclusion of this paper have a certain engineering guiding significance for the pollution discharge assessment and detection of porcelain insulators on site

and provide references for the research of low value and zero value insulators.