Abstract: Common intrusion objects in substations include bird nests, plastic bags, and various types of solid waste. This article proposes a substation intrusion object recognition method based on YOLOv7 for unmanned substations. Firstly, a dataset is created based on intrusion object images captured in the field for preprocessing, including image grayscale, noise addition, data expansion, and data enhancement. Then, LabelImg is used for image annotation to obtain the image dataset containing the data of substation bird nests, plastic bags, and different types of solid waste. Finally, the YOLOv7 model is used to train and test the substation dataset, and the results show the accuracy of YOLOv7 model can reach 96.58% in identifying intrusion objects in substations. The recall rate of YOLOv7 model performs well.