基于Swin-Transformer的岩石自动分类识别

Automatic rock classification and recognition algorithm based on Swin-Transformer

地质勘探领域中,岩石准确识别对于资源评估、勘探定位以及环境保护等方面具有重要意义。然而,传统岩石识别方法依赖于地质学家的观察和经验,存在效率低、主观性强和对专家经验依赖等问题。为了解决以上问题,提出了一种基于Swin-Transformer的岩石自动分类识别算法。该算法通过引入分阶段的注意力机制,将图像分割成不同的块,利用窗口化注意力机制使得每个图像块只与其附近的块进行交互,从而显著降低了计算和内存消耗。实验结果表明,与当前流行的分类模型如ResNet和EfficientNet相比,提出的模型在分类识别效率和准确率上都有显著提高,Top-1准确率达到91.3%,Top-5准确率可达到98.56%。不仅提高了岩石识别的准确性和效率,且通过自动化处理流程,减少了对人工干预的依赖,为地质学研究和工程应用提供了有力支持。

In the field of geological exploration,accurate rock identification is crucial for resource assessment,exploration positioning,and environmental protection.However,traditional rock identification methods rely on the observation and experience of geologists,which suffer from low efficiency,subjectivity,and dependence on expert knowledge.To overcome these challenges,we proposes an automatic rock classification and recognition algorithm based on Swin-Transformer.The algorithm introduces a staged attention mechanism,dividing the image into different blocks and using windowed attention to allow each block to interact only with its neighboring blocks,significantly reducing computational and memory costs.Experimental results demonstrate that compared to popular classification models such as ResNet and EfficientNet,the proposed Swin-Transformer model achieves a significant improvement in classification efficiency and accuracy,with a Top-1 accuracy of 91.3% and a Top-5 accuracy of 98.56%.This research not only enhances the accuracy and efficiency of rock identification but also reduces the reliance on manual intervention by automating the process,providing robust support for geological research and engineering applications.