Precision Agriculture: Enhancing Crops and Weeds Classification of Unbanlance Databases via Weighted-Loss Functions

In the realm of precision agriculture, accurately distinguishing between crops and weeds is essential for optimizing yield and minimizing resource use. This study explores the efficacy of weighted-loss functions in handling unbalanced datasets for image classification in agricultural applications. We conducted experiments under three distinct training scenarios: data augmentation without weightedloss, data augmentation with weighted-loss, and weighted-loss without data augmentation. Our findings reveal that while data augmentation alone yielded an overall accuracy of 97.20%, it failed to address the unbalanced class, resulting in misclassification of the carrot class. Incorporating a weighted-loss function with data augmentation slightly improves the classification accuracy of the unbalanced class, reducing the misclassification error. The most notable improvement was observed when using a weighted-loss function without data augmentation, achieving a validation accuracy of 99.20% and enhancing the unbalanced class accuracy from 66.67% without weighted-loss to 88.89% with weighted-loss. These results underscore the potential of weighted-loss functions in improving model performance on unbalanced agricultural datasets, highlighting their importance in precision agriculture applications.