In the process of power equipment opening/closing X-ray images acquisition by power inspection robot system, there are often low-quality problems such as image distortion, blur and low resolution, which make it difficult to identify the opening/closing state of X-ray images. Therefore, this paper proposes a deep learning recognition method based on data uncertainty learning. Firstly, the recognition algorithm is designed with three convolutional neural networks BaseNet, ResNet18 and MobileNetV3. Then, by fusing the DUL module, the convolutional neural network maps the images space to an uncertain features space that obeys the Gaussian distribution to adaptively learn the noise in low-quality X-ray images. Finally, three groups of comparative experiments are presented to simulate the influence of different quality data on the recognition performance of the model under ideal environment, harsh environment, and normal environment. The experimental results show that the model based on data uncertainty learning performs better than the deterministic model, and the average recognition accuracy of X-ray image opening/closing is increased by 2.64\%. ResNet18+DUL performs best, with an accuracy of up to 100\%, suitable for online recognition. MobileNetV3+DUL is with an accuracy of up to 97.83\%, suitable for offline recognition.