Abstract: The effective image recognition and status analysis of power equipment can improve its operation and reduce the risk of potential power outages in working environment. However, traditional power equipment image recognition methods have problems which are difficult in distinguishing target and background. Apart from this, it is insufficient to extract feature information. An improved attention mechanism network based on image detection and recognition method for power equipment is proposed in this paper. This method proposes a prediction strategy for power equipment feature information: it obtains the prior information of power equipment images through a global structure information learning network module that is rich in detailed features and associated image feature information. At the same time, the learning mechanism of the change process of the depth value is introduced to extract the deep semantic information. Further, the gating mechanism of LSTM (long short-term memory) network is used to predict the feature information of power equipment on different levels of image feature information, constructing an attention mechanism network based on LSTM gating mechanism. In addition, this method proposes a deep and shallow feature information interaction strategy. The matrix outer product method is used to integrate the shallow feature information and deep feature information of the image, so that the model can fully learn the multi-level feature information of the power equipment image. Compared with the traditional power equipment image recognition method, the proposed improvement scheme can strengthen the recognition, accurately distinguish fuzzy boundary information about background and target, enhance the interaction ability of deep and shallow feature information, and improve the accuracy of target recognition in complex background environment effectively. The experimental result shows that the recognition accuracy of the model proposed in this paper reaches 92%, which is higher than that of CvT model.