Hyperspectral remote sensing and machine learning approaches for precise prediction of Ocimum basilicum L. Yield

Résumé fourni par la source

Ocimum basilicum L., an essential oil-bearing medicinal and aromatic crop, is extensively cultivated in India, which produces over 350 Mg of essential oil annually. To meet the demand of this crop, yield estimation prior to harvesting is essential for effective crop management, allowing optimal harvesting to obtain the maximum yield. However, literature indicates a lack of such models for identifying the optimum crop harvesting stage of O. basilicum. Therefore, by leveraging Hyperspectral Remote Sensing (HRS) and Machine Learning (ML) approaches, the present study aimed to develop a precise and robust model for early yield prediction of O. basilicum to achieve maximized crop production. Canopy-level spectral data were recorded using a handheld spectroradiometer sensor throughout the crop growth stages. Several conventional statistical techniques and ML models (Ridge, LASSO, Elastic Net, and Random Forest) were explored. Partial Least Square Regression (PLSR) feature selection model was employed to mitigate the high redundancy in reflectance and its 1st and 2nd order derivatives data. Our analysis revealed that the ridge regression model with PLSR-selected 2nd order derivatives was superior. This yield prediction model achieved R2 = 0.71 in the training dataset and R2 = 0.88 in the testing dataset. The model was applied to field data using hyperspectral imagery, with spectral bands optimized for yield estimation. The predicted yield showed an average difference of 19% compared to actual yields. In conclusion, this study demonstrated the effectiveness of ML-based analysis of HRS data to develop a rapid and efficient method for accurate yield estimation of O. basilicum, prior to harvest.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.