Abstract:Due to the limited availability of ore samples and the issues of overfitting and low classification accuracy in online sorting of ores using deep learning models in the industrial field, a mining data augmentation classification method is proposed that combines X-ray transmission imaging technology. The method is based on an improved Auxiliary Classifier Generative Adversarial Network (ACGAN-gp), which enables accurate prediction of ore grade. Firstly, in terms of the model architecture, the discriminator and generator are constructed using convolutional and continuous residual blocks. An attention mechanism is introduced to capture the detailed features of the ores. The loss function of the discriminator is reconstructed using the Wasserstein distance and gradient penalty to enhance the stability of adversarial training and prevent mode collapse. Secondly, in terms of the algorithm, an auxiliary classifier is added to the training process, which collaborates with the generator to generate high-quality samples and augment the original dataset. Finally, the discriminator is used for classifying the ore samples. Experimental results demonstrate that the proposed method achieves an impressive classification accuracy of 89.62%. The proposed method effectively improves the classification accuracy of ore grades under the condition of limited samples. The generated ore samples by the model exhibit good generalization, and the method can also enhance the performance of other recognition networks in ore grade prediction.