The invention discloses a photovoltaic power station hot spot detection and positioning method based on a thermal infrared image. Pictures shot by a thermal imager carried by an unmanned aerial vehicle are infrared pictures, the resolution ratio is very low, the form of hot spots is usually very small, the accuracy of hot spot recognition is very low, and accurate positioning is difficult. Aimingat hot spot identification, the invention provides a UAV-YOLO model, a neural network residual error unit is added for fusion, characteristics of a front network are utilized to a greater extent, andhot spot characteristics of a far visual angle are learned; a YOLOv3 backbone network architecture is improved, and more hot spot appearance feature information can be learned in a front-end network.Aiming at hot spot positioning, the invention provides a geometric positioning algorithm, and the longitude and latitude of a hot spot are calculated through the longitude and latitude of an image center point, a yaw angle, the shooting height of an unmanned aerial vehicle and other information. According to the invention, the hot spot positioning precision is improved on the premise of ensuring the hot spot identification accuracy.