Infrared image recognition method for nuclear power plant electrical equipment based on EFRM

Artificial intelligence technology has shown significant advantages in the intelligent recognition of electrical equipment in nuclear power plants. A new infrared image recognition method based on efficient feature preservation module and phantom convolution fusion is proposed to address the problems of feature loss and computational redundancy in traditional convolutional neural networks for low resolution infrared image recognition. This method utilizes an efficient feature preservation module to preserve detailed information during the downsampling process, and introduces a coordinate attention mechanism to enhance the representation ability of device spatial position. Meanwhile, reducing redundant feature calculations through phantom convolution. The experimental results show that the proposed method achieves 92.4% in average accuracy, 93.1% in Precision, and 91.6% in Recall metrics, all of which are superior to the comparative methods. In terms of model efficiency, the proposed model has a parameter size of 5.8M, a model size of 23.7 MB, and an inference time of 7.5 ms/sheet, which is also superior to the comparison method. The research results have verified the effectiveness and practicality of this method in infrared image recognition of electrical equipment in nuclear power plants, providing technical support for intelligent operation and safety monitoring.