| Summary: | Remote sensing has a wide range of applications in volcanic geological field investigation, primary for detecting the volcanic lithology types from remote sensing images. However, limited empirical works has explored the effectivity of deep neural network for volcanic lithology classification from remote sensing image. Thus, this paper proposes a new multi-instance feature fusion network, called MIFFNet, as well as considering the image features at different levels, for volcanic lithology classification from remote sensing image. Firstly, the Swin transformer, color moment and gray level co-occurrence matrix are used to extract the high-level semantic features and low-level basic image features of volcanic lithology from VELSD dataset images. Secondly, the two level features are then fused and formed a new synthetic feature of volcanic lithology which greatly enhance the sensitivity and description capability to the volcanic lithology. Finally, the softmax regression is used to further classify volcanic lithology types from remote sensing image. The experimental results show that the MIFFNet becomes more sensitive for volcanic lithology scene, and enabling the network to classify the volcanic lithology type in remote sensing image better. The findings of this paper provide a new perspective and practice for researchers to conduct the volcanic lithology field investigation.
|
|---|
