First Field Validation of a New VNIR–SWIR-Based Six-Band Multi-Camera System for UAVs over Winter Wheat

Abstract

Abstract. Shortwave infrared (SWIR) UAV imaging remains uncommon despite its sensitivity to canopy water and protein. We report, to our knowledge, the first field validation of a six-band, simultaneously exposed VNIR/SWIR multicamera for plot-scale winter wheat. The payload used narrow bandpass filters at 910, 980, 1100, 1200, 1510, and 1650 nm (FWHM 10–12 nm) and was flown at 30 m AGL, yielding 4 cm GSD. Radiometric calibration used in-flight empirical line calibration with an in-scene gray panel set, followed by independent validation on a material-distinct gray set. ASD spectroradiometer measurements were convolved with Gaussian proxy spectral response functions matched to the nominal filter passbands. Empirical line fits were near-perfect (R2 ≈ 1.000; RMSE = 0.003–0.009). Independent panel validation showed near-unity slopes for five bands from 980–1650 nm (R2 = 0.998–0.999; RMSE = 0.005–0.013). Across 36 canopy plot ROIs, camera-to-ASD agreement remained strong for five bands, with slopes of 0.943–1.079, R2 = 0.58–0.85, and RMSE = 0.010–0.023. Two SWIR normalized ratio indices showed tight cross-sensor agreement: NRI[1100,1200] (R2 ≈ 0.93; RMSE ≈ 0.010) and NRI[1650,1510] (R2 ≈ 0.90; RMSE ≈ 0.017–0.018). Post-hoc filter transmittance measurements revealed secondary long-wavelength throughput in the 910 nm channel, causing compressed slopes and elevated error (MAPE ≈ 33%); this band was excluded from accuracy claims. Panel-anchored, bandpass-aware calibration enables quantitative UAV SWIR reflectance and robust SWIR indices for precision agriculture applications. The workflow also identifies hardware-specific failure modes, supporting reproducible validation and informed redesign of filter-reconfigurable SWIR payloads.

Code and data availability

The supplied article blocks describe a UAV VNIR–SWIR multicamera validation over winter wheat, but contain no data availability, code availability, or repository statements. No public phenotype datasets, imagery, analysis code, models, or supplements with authors' URLs are mentioned; only the article DOI itself appears

No evidence-backed public reproduction asset is currently recorded.