Classifying automotive parts during manufacturing is crucial for effective supply chain management and quality assurance. This paper proposes a grayscale image classification method for auto parts utilizing an enhanced MobileNetV3_Small model. Augment the original data set with data enhancement techniques. The spatial attention mechanism is added before the original SE attention mechanism to enhance the feature extraction capability of the MobileNetV3_Small network on grayscale images. Extensive experiments conducted on ten types of balanced datasets demonstrate that the model achieves an accuracy of 99.4%, with a loss rate of only 0.0054%. Furthermore, metrics including precision, recall, F1 score, and AUC all reach perfect scores of 1.00. These results comprehensively validate our proposed model’s high accuracy and efficiency while showcasing its robust classification capabilities. Notably, the model's weight file is merely 6.2 MB in size, facilitating rapid inference and efficient deployment, meeting the industry demands for rapid classification of parts, and highlighting its advantages in adaptability for practical applications.