Homogeneização radiométrica de imagens RGB de VANT

Abstract

Radiometric inconsistency among Unmanned Aerial Vehicle (UAV) images acquired at different times of day compromises the comparability of remote sensing products applied to precision agriculture. This study evaluated the capacity of the Histogram Matching algorithm to promote relative radiometric homogenization of UAV RGB images over a coffee crop, comparing scenes captured at 09:00 and 13:00 with a reference image acquired near solar zenith (11:00). Based on 57 sampling points, descriptive statistics, Pearson correlation, Lin's Concordance Correlation Coefficient (CCC), Bland-Altman analysis, and a paired Wilcoxon test were computed for the three spectral bands. Results showed that equalization reduced point-to-point dispersion by 52% to 65% across all bands and times, and raised the CCC of the red and green bands from the "poor" to the "substantial" to "almost perfect" range, with a statistically significant reduction in median absolute error (p < 0.01). Pearson's coefficient also increased markedly after correction, indicating that the method, being monotonic but non-linear, does not preserve Pearson correlation. In the blue band, Pearson increased similarly, but CCC remained modest and no statistically significant improvement was found, revealing a residual systematic bias. It is concluded that Histogram Matching provides effective radiometric correction for the red and green bands, but fails to correct the systematic bias of the blue band, possibly related to diffuse atmospheric scattering, which is more intense at that wavelength.