The invention relates to the technical field of artificial intelligence and power transmission and transformation, in particular to a substation foreign matter recognition method, device and system based on deep learning. The method comprises the following steps: acquiring a scene image of a substation; recognizing foreign matters in the scene image according to a pre-established substation foreign matter target detection model to obtain the positions of the foreign matters in the scene image and the types of the foreign matters; wherein the substation foreign matter target detection model isobtained by training according to a preset foreign matter type in a substation and a plurality of substation foreign matter images. According to the invention, the foreign matter detection accuracy isimproved, the types of foreign matters can be specifically distinguished, and the inspection and monitoring of the substation can be assisted. Modules developed by the method can be integrated to a fixed camera, a robot body and remote control systems of the fixed camera and the robot body, the fixed camera and the robot body are endowed with the capability of intelligently identifying foreign matters in a transformer substation, and the intelligent level of substation inspection and monitoring is improved.