The use of remote sensing technology for geological and mineral resource assessment is becoming increasingly attractive and cost-effective due to its ability to provide a more accurate understanding of lithology and structural information. This study leveraged remote sensing datasets, feature-level fusion, a U-Net Convolutional Neural Network (U-Net CNN), and fieldwork to address issues in lithological mapping in humid tropical regions. The autonomous district of Yamoussoukro, situated in central Côte d’Ivoire, is characterized by a generally hot and humid climate and is recognized as a suitable experimental area for extracting remote ground information. Based on the Sentinel satellite’s optical and radar datasets, we applied several false-color composites (FCCs), Band Ratios (BR), and Principal Component Analysis (PCA) to pinpoint six lithological feature maps. Such derived features, including the volcanic rock index, granitoid facies index, sedimentary rock index, principal components, and shortwave bands, were fused using stacked spectral index (SSIF) fusion techniques to create a new multispectral image that incorporates key lithological features. Subsequently, a U-Net-based semantic segmentation was applied to the hybrid multi-channel image. The model yielded an overall accuracy of 95.38%, a Mean Intersection over Union (mIoU) of 81.26%, and improved the extraction of granitoid groups and rocks from the volcano-sedimentary formations of the Yamoussoukro district. The resulting lithological map serves as an indicator for mineral exploration and contributes to establishing the district’s geological resource management system. • Lithological mapping through U-Net Convolutional Network model • Features fusion through Stacked Spectral Indices (SSIF) Fusion technique • Lithological features extraction using FCCs, BR, and PC Analysis • Field campaign based on specific checkpoints to confirm the lithological categories
Kouame et al. (Sun,) studied this question.