The malfunction of power equipment may lead to instability or even disconnection of the power system, causing huge losses to power safety and the national economy. It is crucial to quickly and accurately identify these malfunctions. Infrared image features exhibit excellent feature expression ability in capturing the heating malfunctions of power equipment. However, during the image acquisition process, problems such as object overlap, occlusion and interference between categories may occur. Therefore, a complex image fault recognition algorithm based on multilayer deep neural networks is proposed in this paper, fully utilizing the high-level feature extraction capability of multilayer network modules and the feature fusion capability of multilevel network modules to improve the accuracy of fault recognition. Experimental results show that the proposed algorithm is superior to existing models such as Faster-RCNN, VGG16, VGG19 and traditional Resnet models in terms of accuracy and running time, and its effectiveness in solving problems such as target overlap,occlusion and dass interference in images is verified.