Abstract: Aiming at the low recognition accuracy and training efficiency of neural network for photovoltaic infrared thermal image in photovoltaic generation,a recognition model based on scatter plot-AlexNet network is proposed. First,the HSI spatial data information corresponding to the photovoltaic infrared thermal image is directly projected into the plane coordinate system to form a visual RGB three-color image composed of several scattered points,and the rapid AlexNet network based on the AlexNet network is used for training. In order to prove the excellent performance of the new method,the recognition accuracy and single training time are selected to evaluate the model. Experimental results show that this method has high recognition rate,can accurately identify the types of images in photovoltaic infrared thermal images,and the training efficiency is significantly improved.