A Classification Migration Model of Rock Images Fusing SE-Net

WANG Zhiwei; TAN Meilin; GAO Xianjun

doi:10.3969/j.issn.1672-9722.2025.03.015

Abstract：With the continuous development of deep learning technology,convolutional neural networks provide new solutions for rock image classification.However,the weak feature discrimination between different rock images results in the model's classifi-cation accuracy and speed cannot meet the needs of practical applications,so a rock image classification migration model fused with SE-Net(Squeeze-and-Excitation Networks)is proposed.The model is based on the VGG(Oxford Visual Geometry Group)convo-lutional neural network model,and embedded SE-Net before its fully connected layer,focusing on the distinctive feature regions in the rock image.Then,through the network structure of the ImageNet data set Migration learning is carried out on the basis of param-eters,and a deep learning migration model incorporating the attention mechanism is constructed to enable it to capture the multi-level features of the rock faster and more accurately,thereby realizing rapid and accurate classification of rock images and greatly improving the classification accuracy of rock images and efficiency.The network model is tested using the collected seven types of rock images,and the test results show that the accuracy rate reaches 94.12%.Compared with the existing networks VGG16,ResNet50 and their migration models,the training loss,verification loss and model parameters are all obvious reduce.Experiments show that the proposed network model can effectively improve the classification accuracy of rock images under the premise of ensur-ing a small amount of model parameters and low requirements for experimental machine configuration.