Answers are grounded in this paper and cite their sources.
High-resolution remote sensing farmland change detection is crucial for agricultural resource management. However, existing methods still face challenges such as confusing linear features with farmland boundary texture, boundary shift due to downsampling, and deep feature edge structure degradation when accurately inscribing change boundaries. To this end, this paper proposes a farmland change detection network (AMDDSMNet) based on asymmetric multidirectional convolutional attention and dynamic spatial modulation. The network first suppresses interference and irrelevant changes through spatial-spectral dual-domain difference modelling; subsequently, it introduces a library of learnable primitives via a Dynamic Spatial Modulator (DSM) to incorporate topological priors, thereby significantly enhancing edge continuity. Concurrently, an asymmetric multi-directional convolutional attention (AMCA) module is employed to dynamically focus on multi-scale linear boundaries via a directional attention mechanism, effectively enhancing edge response sensitivity. Comprehensive experiments on three farmland-specific datasets, CLCD, GFSWCLCD, and LJCLCD, and the building change dataset, HRSCU-CD, show that, for the farmland change detection task, AMD-DSMNet achieves the optimal performance on all the dedicated datasets, with F1 scores improved by$0.13%$-$1.32%$compared to the state-of-the-art algorithms. The IoU enhanced by$0.18%$-$1.68%$. This study effectively solves the problem of accurate boundary localization for farmland change detection.
No takes yet. Share an insight, caveat, or question.
Yang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: