Lithological Remote Sensing Scene Classification Based on Multi-view Data

Lithology classification is an important branch of remote sensing of geological environment. Deep learning method has strong feature extraction ability and has been widely used in the field of geological remote sensing classification. However, in vegetated areas, the features of remote sensing lithology images are complex, and it is difficult to effectively extract the key feature information of lithology. To solve these problems, based on deep learning, this study carries out research from the two core levels of data and model. To eliminate the boundary effects caused by the interference of feature information of multiple types of lithology, a method of specifying the label of the scene graph after cropping was proposed, which can effectively eliminate the multi-level cropping at the boundary position. This work provides data support for subsequent model training. In order to improve the ability of the model to extract key information of lithology, a lithologic scene classification network model based on enhanced feature fusion and channel attention (EFFCA) was proposed by using the dense connection network and channel attention mechanism. Then based on the strategy of feature-level fusion and data-level fusion, we use EFFCA to construct a multi-view data fusion lithological scene classification model. Experiments on self-constructed multi-view lithology datasets show that compared with VGG16, DenseNet121 and other models, our proposed model achieved better performance. The results of this study can provide theoretical method support for scene classification of lithological remote sensing data, and have certain scientific significance and application value.