Satellite Image-Based Land Use and Land Cover Classification Using Spectral-Spatial Consistency Guided Hybrid Convolutional Neural Network-Transformer Network
Year: 2026
Type: conference-paper
Abstract: The increasing accessibility of high-resolution satellite imagery has created significant opportunities for automated LULC classification in environmental monitoring, agricultural management, urban planning and natural resource assessment. However, existing classification approaches often struggle to simultaneously capture fine-grained spectral-spatial patterns and long-range contextual dependencies for visually similar land-cover categories. To overcome these limitations, a Spectral-Spatial Consistency Guided Hybrid Convolutional Neural Network -Transformer Network (SSCG-CTNet) for accurate satellite image-based LULC classification. The proposed framework integrates Spatial feature learning using CNNs with Transformer-based contextual modeling to learn complementary spatial and spectral representations. Furthermore, a novel Spectral-Spatial Consistency Principle (SSCP) is introduced for enhancing the classification reliability by jointly evaluating spectral evidence and spatial contextual relationships. The consistency-guided classification mechanism refines class probabilities while encouraging spatially coherent predictions and preserving discriminative spectral characteristics. The proposed framework is designed to improve robustness against intra-class variations, inter-class similarities and complex landscape structures. Consequently, SSCG-CTNet provides an efficiency and reliable solution for automated LULC classification and large-scale satellite image analysis.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7225083007