Abstract: Aiming at the problems existing in the photovoltaic hot spot recognition algorithm,such as the complex calculation of deep network parameters,the easy disappearance of gradient information and the reduced accuracy of model degradation,a photovoltaic hot spot identification and detection algorithm based on feature pyramid fusion high-resolution network is proposed. Firstly,a network model with parallel connection of multi-resolution subnetworks is build in this algorithm,which solves the problems of loss of detailed information and redundant features of hot spots in deep networks. Secondly,the multi-scale fusion module of the feature pyramid is introduced,which connects the feature maps of different scales across layers,solves the feature semantic gap,and improves the accuracy of model recognition. The experimental results show that the classification effect of the proposed algorithm on the photovoltaic infrared hot spot image dataset is better than the classical deep convolutional neural network algorithm,and the accuracy rate can reach97.2%. The algorithm realizes high-precision and high-resolution hot spot detection and identification.