MULTI VEGETABLE LEAF DISEASE CLASSIFICATION AND FERTILIZER RECOMMENDATION USING OPTIMIZED DEEP LEARNING MODEL WITH SHAPLEY ADDITIVE EXPLANATIONS

Abstract

Multi-vegetable leaf disease classification and fertilizer recommendation efforts on identifying and classifying diseases in vegetable leaves using advanced image processing and machine learning techniques. Moreover, several existing approaches faced major challenges such as low accuracy, complexity issues, scalability limitations, and efficiency problems. To overcome the aforementioned issues, this research proposes a very new Dense Scale-invariant feature transform-based Orientation-guided Convolutional Attention network with Quokka Swarm Optimization (DSOCA-QSO) and further adaptation of QSO towards multi-class disease classification. This approach achieves high accuracy with minimal training data, optimizing agricultural productivity through targeted fertilizer recommendations. The process in the pre-processing pipeline includes resizing, normalizing, denoised, and contrast-enhanced by the AEFGC which enhances the quality of images as well as noise-free images. The DSOCA network employs orientation-guided convolutional attention extraction for feature extraction, which allows the model to direct attention to important patterns for diagnosis purposes. In contrast, the QSO model fine-tunes the weight parameters of the model, which it uses to address the problems of computational complexity and sensitivity to parameter values. Additionally, SHAP is employed to enhance interpretability by explaining the model’s predictions, ensuring transparency. Fertilizer recommendations are tailored to specific soil conditions, promoting plant health. The DSOCA-QSO approach achieves accuracy (99.9%), F1-score (99.35%), precision (99.7%), recall (99.99%), and sensitivity (98.99%). These results highlight its effectiveness in improving agricultural practices through precise disease diagnosis and optimized fertilizer recommendations.