Bi-temporal semantic change detection of rice disease and lodging based on difference token transformers and cross-branch distillation
Year: 2026
Type: article
Abstract: Semantic Change Detection (SCD) in high-resolution remote sensing imagery is essential for precision agriculture, particularly for timely monitoring of crop stresses. Recent studies have demonstrated the effectiveness of constructing SCDs using a triple-branch convolutional neural network (CNN), which includes two temporal branches and one change branch. However, existing methods struggle to distinguish progressive pathological developments from normal phenological growth. Therefore, in this paper, a transformer-based temporal difference modeling module is introduced to explicitly capture bi-temporal semantic discrepancies via difference tokens, enabling effective modeling of subtle crop condition changes. To further enhance cross-task consistency, a cross-branch distillation strategy is developed to facilitate bidirectional knowledge transfer between semantic segmentation and change detection branches, thereby improving feature discriminability and training robustness. Extensive experiments on high-resolution agricultural remote sensing datasets validate the effectiveness of the proposed framework, which achieves an mIoU of 81.90%, SeK of 69.39%, Acc of 91.22%, and an F csd of 93.54%, outperforming state-of-the-art semantic change detection methods. The proposed approach provides a reliable solution for crop growth monitoring, disease-related change analysis, and lodging assessment in precision agriculture.
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
SDG: Zero hunger
Open Access status: gold
Estimated APC paid: $1,100
OpenAlex ID: W7225080353