Visible light image based insulator recognition is one of the most important tasks of intelligent inspection equipmet identification in power grid. Generally, the samples captured from intelligent inspection in power grid are of low availability and contain multi-scenes. These problems greatly limit the application of deep learning in intelligent recognition of power grid equipment. In this paper, a small number of multi-source images are explored and applied in insulator recognition based on deep learning. Firstly, the development process of object recognition algorithm based on deep learning is described. Secondly, three deep learning models for object recognition are emphatically introduced and compared,which are region-convolutional neural network (R-CNN), Fast region-convolutional neural network (Fast R-CNN) and Faster region-convolutional neural network (Faster R-CNN). Finally, through experiments on a hundred magnitude insulator images from different scenes, the usability and robustness of the Faster R-CNN model are verified. The research and experiment in this paper have explored an effective way for the application of deep learning technology in the object recognition of various equipment in power grid.