A Deep Learning Approach Combining GLCM Features with UNet++ and GNN for Salt Body Detection in Seismic Data

Accurate detection and segmentation of salt bodies in seismic data is critical for effective subsurface interpretation and hydrocarbon exploration. Traditional methods often struggle with complex geological patterns, motivating the use of deep learning for enhanced accuracy. In this ‎study, we propose a hybrid deep learning framework that integrates Gray Level Co-occurrence Matrix (GLCM) texture features with the ‎UNet++ architecture and Graph Neural Networks (GNNs) to improve salt body detection. The model combines handcrafted texture ‎descriptors with deep hierarchical features and spatial relationships, offering a richer understanding of salt structures. The publicly available ‎TGS Salt Identification Challenge dataset is employed for training and evaluation. Preprocessing steps include denoising, contrast enhance-‎ment, and grayscale normalization, while GLCM is used to extract statistical texture patterns. These features are fused with encoder outputs ‎in UNet++ and further refined using GNN layers to exploit inter-pixel dependencies. Performance is evaluated using metrics such as Intersection over Union (IoU), Dice coefficient, Precision, Recall, and Accuracy. Experimental results demonstrate that the proposed approach ‎significantly outperforms baseline UNet and CNN models, especially in edge delineation and salt boundary preservation. This framework ‎holds promise for practical seismic interpretation workflows, particularly in complex stratigraphic environments‎.