Estimation of Photosynthetic Growth Signature at the Canopy Scale Using New Genetic Algorithm-Modified Visible Band Triangular Greenness Index

Greenness index has been proven sensitive to vegetation properties for multispectral and hyperspectral imaging. However, most controlled microclimatic cultivation chambers are equipped with low-cost RGB camera for crop growth monitoring. The lack of camera credentials specially the wavelength sensitivity of visible band provides added challenge in materializing greenness index. The proposed method in this study compensates the unavailability of generic camera peak wavelength sensitivities by employing genetic algorithm (GA) to derive a visible band triangular greenness index (TGI) based on green waveband signal normalized TGI model called gvTGI. The selection, mutation and crossover rates used in configuring the GA model are 0.2, 0.01 and 0.8 respectively. Lettuce images are captured from an aquaponic cultivation chamber for 6-week crop life cycle. The annotated and extracted gvTGI channels are inputted to deep learning models of MobileNetV2, ResNetl01 and InceptionResNetV2 for estimation of photosynthetic growth signatures at canopy scale. In predicting cultivation period in weeks after germination, MobileNetV2 bested other image classification models with accuracy of 80.56%. In estimating canopy area, MobileNetV2 bested other image regression models with $\mathrm{R}^{2}$ of 0.9805. The proposed gvTGI proved to be highly accurate on estimation of photosynthetic growth signatures by using generic RGB camera, thus, providing a low-cost alternative for crop phenotyping.