Under ideal illumination conditions, the existing face recognition algorithms can obtain better recognition results. However, under non-uniform illumination conditions, the performance of the face recognition algorithm will be greatly reduced. Moreover, the common facial image classification models have low precision and speed. To address these problems, this paper proposes a non-uniform illumination face recognition method based on the improved MobileNetV1, and constructs a lightweight feature extraction network with MobileNetV1 as the core. To reduce the influence of non-uniform illumination on the face image, we use the MSRCR algorithm to preprocess the face image. A lightweight I-MobileNet model is proposed. We introduce the attention module into the MobileNetV1 model, which enhances the feature extraction capability of the network and improves the recognition performance. At the same time, the model parameters are optimized with Arcface loss function to increase the distance between classes and reduce the distance within classes. Compared with the original MobileNetV1 model, our method shows a significant improvement in recognition accuracy. In comparison with other network models, experiments conducted on the Extended Yale B dataset with the I-MobileNet model demonstrate the effectiveness of the method.