Failures of the photovoltaic (PV) module can affect the performance of the PV arrays, thus reducing the power generation efficiency. In serious cases, PV module failures may even jeopardize the safe operation of the power plant. Traditional methods cannot meet the current demand for fast and correct PV module fault detection. Therefore, this paper proposes a PV module fault identification method based on the improved EfficientNet algorithm. First, the collected infrared images of PV modules are utilized to establish a fault dataset, which is then preprocessed by using image segmentation and data enhancement technology. Second, a fault recognition model is constructed based on the EfficientNet network. Meanwhile, a dual-channel convolutional block attention module (CBAM) is introduced into the model, which can inhibit the recognition of unnecessary features and enhance the ability to extract spatial feature information, thus improving the recognition accuracy. Finally, comparative simulation experiments are conducted to validate the effectiveness and advancement of the proposed model. The experimental results show that the fault recognition accuracy of the model reaches 90. 83% , which is 2. 83% higher than that of the original EfficientNet model; in addition, the model size is only 20. 3 M, which shows good practicability and can meet the requirements of practical application of PV power plants.