Precision Nutrition Management and Fertilizer Optimization in Paddy Crops: A Hybrid Approach for Deficiency Detection and Recommendation Using Segmentation, Transfer Learning and Hyperparameter Tuning

Abstract

The accurate and rapid diagnosis of nutrient deficiencies in crops is essential for effective agricultural management. Conventional methods, which rely on visual inspection of crop symptoms are limited by subjectivity and require significant expertise, making them impractical for widespread use by farmers. In this study, we propose a novel approach utilizing digital imaging and deep learning to quantitatively analyze crop symptoms for nutrient deficiencies, specifically targeting paddy crops. Our methodology involves segmenting the foreground of the crop image by removing background noise using GrabCut, refining the input image to enhance clarity and improve the accuracy of nutrient deficiency detection. We introduce a novel approach combining deep learning architectures, specifically MobileNet and a fine-tuned variant of MobileNet, for nutrient deficiency classification. To address overfitting, we integrate dropout layers and optimize hyperparameters, including learning rates and optimizers. The performance of these models is assessed using established metrics, including accuracy, precision, recall, and F1 score. Notably, the base MobileNet model achieves an accuracy of 89.65%, while the fine-tuned MobileNet variant attains 93.10%, demonstrating significant improvement and superiority. This integrated approach presents a promising solution for efficient nutrient management in paddy cultivation, contributing to increased yields and sustainable agricultural practices. Additionally, our system recommends appropriate fertilizers based on the nutrient deficiency findings, augmenting precision agriculture and crop management practices.