Abstract: Aiming at the problems of low contrast,complex background and difficult detection of small hot spots in aerial photovoltaic infrared images,a cross-stage partial network incorporating Swin transformer（CSwin）is proposed to reduce the number of parameters and capture the global position and spatial information of the image. The multi-scale feature path aggregation network based on CSwin（MPC）is constructed to strengthen the information interaction of multi-scale features and further improve the small target detection ability. The qualitative and quantitative experiments are carried out on the aerial photovoltaic infrared image data set, and the effectiveness of the method in aerial photovoltaic infrared image hot spot detection in proved.