Rock image classification based on improved EfficientNet

Rock image classification plays a crucial role in geological exploration, mineral resource development, and environmental monitoring. However, rock images often exhibit high intra-class similarity and low inter-class variation, posing challenges for accurate classification. Additionally, existing models often suffer from having a large number of parameters. To address these issues, we propose an enhanced rock classification model based on EfficientNet-B0. First, the DiffuseMix algorithm is applied to the training set to increase data diversity. Second, the shuffle attention mechanism is integrated into the backbone network to enhance feature extraction while reducing model parameters. Finally, the Lion optimizer is employed to optimize the training process, improving both the accuracy and stability of the model. The experimental results demonstrate that the improved model achieves an accuracy of 94.02% on the test set, outperforming ResNet50, MobileNetV3, and SwinTransformer by 8.02%, 9.35%, and 5.02%, respectively. Additionally, the model's parameter count is significantly reduced to 3.38 million. The proposed model reduces the number of parameters while maintaining high recognition accuracy for rock image classification, providing a novel solution for intelligent rock recognition.