An Efficient Machine Learning Enabled Non-Destructive Technique for Monitoring of Crop Remotely

Crop health can be predicted based on various biochemical variables of crops, which include chlorophyll, phenol, carbohydrate, lipid, protein, hydrogen peroxide, and proline as these variables play a critical role in maintaining the intricate phytochemistry of crop plants. In-situ monitoring of the above-mentioned variables is very cumbersome and laborious, so it is atrociously needed to identify some alternatives to monitor these variables in crop plants. Assessing these variables using satellite data may be a good choice provided it has a high spatial and temporal resolution. Sentinel-2 satellite sensor contains VNIR/SWIR spectral region, two red-edge bands, and also has good spatial as well as temporal resolution so it may be the finest option. Precise information of the field is required for the development of a retrieval algorithm for which the drone data is used, as it is highly accurate ground truth reference data. In previous studies, the researchers have primarily focused on the monitoring of biophysical and morphological parameters of crop plants like leaf area index, plant height, and stomatal conductance using these spectral features. Because monitoring biochemical variables of crop plants using satellite derivatives is still a difficult undertaking for academics, just a few studies have been published. As a result, in this study, an attempt is made to establish a methodology for monitoring sugarcane crop biochemical characteristics utilising satellite-derived variables. Satellite derivatives, i.e., vegetation indices are extracted using Sentinel-2 data while biochemical variables of the crop (as mentioned above) are analyzed using leaf samples in the laboratory using optimized protocols. Subsequently, a Machine learning-based Gaussian process regression model is developed for all the biochemical variables using the different combinations of vegetation indices. The developed model showed promising results with R2 greater than 0.7 and NRMSE less than 0.2 thus holding good potential for effective monitoring the crop health condition remotely.