Effective management of rye cover crops in cash-crop systems relies heavily on accurate biomass estimation. Low-altitude Unmanned Aerial Vehicle (UAV) imagery offers a promising high-resolution alternative, yet unlocking its full potential requires moving beyond basic estimation models to more integrative and explanatory models. This study obtains the measured height (MH), SPAD (Soil and Plant Analyzer Development) values, and measured dry biomass (MDB) and applies UAV remote sensing and machine learning to acquire the crop canopy height, vegetation indices (VIs), and vegetation fraction (VF) across growth stages. Among single-parameter biomass estimation models, the estimated height yields the best at the overall growth stage (R2 = 0.935), whereas selected VIs perform the best at the non-seedling stage (R2 = 0.851). For multi-parameters modeling, models combining height, VF, and VIs significantly outperform the single-parameter models, achieving better estimation results throughout each growth stage (Best R2 = 0.951). Structural equation modeling clarifies the direct and indirect contributions of these parameters to biomass accumulation, revealing their synergistic effects. This study demonstrates the potential of UAV-based multi-parameter biomass estimation model to support more informed decisions in cover crop management and to advance broader precise agriculture practices. Additionally, the analytical framework developed here offers a transferable approach for high-resolution biomass monitoring in other crop systems.
Delineating early cover-crop vegetation in unmanned aerial vehicle (UAV) imagery requires preserving slender leaves amid cast shadows, desiccated residue, and heterogeneous soil. Weak foliage contrast and leaf-like background edges make appearance and geometry jointly important. We propose the Shadow, Topology, and Chromatic-aware Prior Feature Topology Fusion Network (STC-PFTF-Net), which integrates structure-guided reconstruction with complementary appearance representations. A coherence-guided tangent–normal decomposition preserves tangential sampling displacement and attenuates displacement across local structures. Chromatic–photometric encoding, directional wavelet filtering, and local connectivity refinement support this reconstruction throughout the feature hierarchy. On the 32 validation tiles of the Residue-rich Cover-Crop UAV Dataset, drawn from an area spatially separated from training, STC-PFTF-Net achieves 77.71% foreground intersection over union (IoU), 50.88% Boundary IoU, 83.84% centerline Dice (clDice), and 87.35% F1. At a common decision threshold, the complete model also exceeds the comparison architectures in all four measures. Spatial analyses associate its higher boundary recall with vegetation–background transitions and identify low-greenness foliage as a distinct source of omission. These findings show the value of combining appearance information with explicit reconstruction geometry for fine-scale canopy delineation. With 2.12 million parameters, the model offers a compact representation for UAV-derived vegetation mapping and subsequent canopy-cover assessment.
Road incidents involving public-transport vehicles and hazardous-material carriers require ground-based RGB sensing capable of localizing full vehicle extent while identifying visible fire cues under severe background clutter, scale variation, and illumination artifacts. To address the tension between preserving macro-level vehicle semantics and capturing localized fire patterns, we propose the Robust Dual-Path Fusion Detector (RDF-Det), centered on a parallel feature-fusion neck (RDF-Neck) that integrates top–down semantic propagation with bidirectional spatial aggregation. By decoupling the dual-path topology from branch-mixing parameterization, we systematically investigate four routing strategies—level-wise learned, globally shared, fixed-uniform, and parameter-matched frozen routing—and evaluate cross-family instantiations across RTMDet, YOLOX, and Faster R-CNN. Across a 100-run five-seed benchmark, the main RDF-Det configuration (RTM-RDF-GLR-DINO) achieves mean validation scores of 0.6352 AP and 0.9290 AP50, delivering a system-level improvement of 7.24 AP points over RTM-Base and outperforming the YOLO26 detector (0.6235 AP and 0.9071 AP50). Controlled evaluations on both the validation split and a screened test subset demonstrate that lightweight global and fixed-uniform fusion remain highly competitive with complex level-wise routing, while cross-family comparisons clarify the architectural conditions governing one-stage and two-stage adaptation. Supported by repeated runs, confidence intervals, and Holm-corrected paired tests, this study provides both an effective vehicle-fire detection framework and systematic empirical insights into dual-path multi-scale fusion.