Answers are grounded in this paper and cite their sources.
Food security and sustainable agriculture require accurate, large-scale crop type mapping (CTM), yet traditional methods remain labor-intensive and nonscalable. This article presents an end-to-end artificial intelligence approach for automated CTM. Our key contributions include: first, A novel five-channel ConvNet (ResNet backbone) parcel segmentation model combining NICFI RGB, digital elevation model, and slope data, which outperforms mean shift clustering and, in a channel ablation, improves average precision ( $+$ 5.8%) and IoU ( $+$ 9.9%) over the RGB-only baseline, while reducing false-positive parcel detections in mountainous terrain, with inference time reduced by a factor of three (4 h versus 12 h for state-wide processing). Second, A scalable crop classification approach using time-series models (TST, InceptionTime) applied to S1 (VV/VH) and S2 (B4, B6, B7, B12) data. By integrating our ConvNet segmentation, classification accuracy improves by 3% (81% overall accuracy) with F1-scores ≥ 88% for major crops (sugarcane, agave, avocado). Third, practical solutions for real-world challenges, including: weighted loss functions to handle highly imbalanced datasets, parcel-based analysis to mitigate noisy labels and improve scalability, and multiseason validation demonstrating stability for perennial crops (e.g., Agave). The system’s robustness is further validated via external label comparisons and state-wide deployment across Jalisco. To the best of authors’ knowledge, this is the first work to combine multisensor satellite time-series data with five-channel ConvNet segmentation for large-scale CTM.
No takes yet. Share an insight, caveat, or question.
Sánchez-Pérez et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: