Abstract: To address the problems of insufficient recognition accuracy for small-scale defects and poor real-time performance of photovoltaic (PV) hot-spot detection methods in a complex environment, this paper proposes the deep learning models based detection method for PV hotspot to improve the detection accuracy and speed of PV panel hotspot faults, which provides support for operations of PV power stations. Through YOLOv8 and YOLOv11 algorithms, the problems of hotspot image recognition tasks are addressed. The experimental dataset of 6000 infrared images of PV panels covering typical hotspot faults is established. Through model training and field deployment verification, the results show that in the small object detection task of PV module hotspot fault detection, the average precision of the YOLOv8 and YOLOv11 models on the test set reach above 94% and 99% respectively. Specifically, YOLOv11n improves by 7.1% and 8.8% compared to YOLOv5n and YOLOv9t models. Experiments on YOLOv11 with different model sizes demonstrate that YOLOv11 outperforms the YOLOv8 model in terms of model size and inference speed. This YOLOv11 deep learning model is suitable for detecting low-resolution hotspot fault images and other small object detection scenarios.