【摘要】 岩心岩性是反映地质条件的重要指标，传统的岩性鉴定通常依赖于人工目视检查，既费时又对专业水平要求高。近年来，卷积神经网络技术的迅速进步，为岩心图像的自动化预测开辟了一条新的途径。本文提出了一种基于改进的ResNet50网络结构的岩心图像分类算法，通过引入ECA(Efficient Channel Attention)注意力机制和PSA (Pyramid Scene Attention)注意力机制，改善了网络对岩心图像丰富地质信息的提取和理解能力，对提高岩性分类的准确性和客观性起到了重要作用；引入可变形卷积(DCNv2),使模型能够自动适应图像特征不规则性和形状变化，显著提升了对岩心结构复杂性的识别能力；使用迁移学习方法，提高了模型的泛化能力和训练效率。实验结果表明，改进的ResNet50网络模型在岩心图像分类任务上表现优异，相较于其他主流卷积网络，平均准确率明显提升，较基线网络ResNet50提高了2.33%的准确率，也有效地提高了对复杂岩心结构的识别精度与鲁棒性。更多还原

【Abstract】 The lithology of rock cores serves as a crucial indicator reflecting geological conditions. Traditionally, lithological identification relies heavily on manual visual inspection, which is both time-consuming and requires a high level of expertise. In recent years, the rapid development of CNNs(Convolutional Neural Networks) has provided an innovative approach for automated prediction of rock core images. This paper proposes an improved ResNet50 network-based algorithm for rock core image classification. By incorporating ECA(Efficient Channel Attention) and a specifically designed PSA(Pyramid Scene Attention) mechanism, the algorithm enhances the network′s ability to extract and comprehend the rich geological information present in rock core images, thereby playing a pivotal role in improving the accuracy and objectivity of lithological classification. Furthermore, the introduction of DCNv2(Deformable Convolution Networks version 2) enables the model to automatically adapt to irregularities and shape variations in image features, significantly enhancing its recognition capabilities for the complexity of rock core structures. Additionally, utilizing transfer learning methods improves the model′s generalization ability and training efficiency. Experimental results demonstrate that the modified ResNet50 network model excels in the task of rock core image classification, achieving a notable increase in average accuracy compared to other mainstream CNNs. Specifically, it outperforms the baseline ResNet50 model by 2.33% in accuracy, effectively elevating the recognition precision and robustness for complex rock core structures.更多还原