MACHINE LEARNING AND REMOTE SENSING INTEGRATION FOR EARLY DETECTION OF CROP DISEASES AND PEST OUTBREAKS

Abstract

 This study examines how machine learning algorithms can be used to combine with remote sensing technologies to detect agricultural diseases and insects infestations on a timely basis.  Our mixed-method dataset consisted of an integration of multispectral and hyperspectral images with field data and farmer interviews (ground-truth data).  The steps preceding it were the correction of the atmosphere, normalization of reflectance and extraction of features based on vegetation indices and texture measurements.  Training and testing multiple machine learning models, including support vectors machine, gradient boosting, convolutional neural network, and random forest, were performed with a stratified k-fold cross-validation approach.  These models demonstrated that they were able to give correct predictions with ROC-AUC values that were greater than the baseline values and F1-scores demonstrating that they were able to be able to detect damaged crops.  The models were validated with field surveys and qualitative farmer feedback and ensured their usefulness in real-world applications in agriculture.  The findings revealed that combination of both quantitative spectral attributes and qualitative observations reduced the rate of misclassifications and enabled the system to be more stable in the evolving environments.  The integrated process reveals a scaling, sustainable, and adaptable approach to performing precision agriculture that will be capable of prompt reactions that can reduce agricultural wasteage and enable food security to be more robust.  The findings indicate that machine learning and remote sensing may be combined to establish an alternate method of real-time monitoring of a disease and pests.