Adaptive Multi-Stage Deep Learning Framework for Real-Time Banana Leaf Disease Diagnosis in Smart Agriculture

Abstract

Banana crops are highly susceptible to Black Sigatoka, Panama Disease, Moko, and Yellow Sigatoka that causes significant yield reduction if not identified at an early stage. Manual visual inspection by specialists and subjective to time makes traditional disease diagnosis slow and inappropriate in large-scale agricultural settings. This paper proposes an automated banana leaf disease detection framework based on machine learning and deep learning and computer vision techniques for smart agriculture to support smart agriculture. The suggested system utilizes multi-stage processing pipeline, which includes image acquisition, preprocessing, feature learning and disease classification. Two classification models are applied and compared with each other: Support Vector Machine (SVM) as a standard machine learning model and Convolutional Neural Networks (CNN) as a deep learning model that can be programmed to automatically identify discriminative visual characteristics of leaf images. The framework is explained to describe seven banana leaf conditions, which include various types of disease and healthy leaves. Experiments are conducted using a publicly available image dataset of banana leaves and performance is measured using accuracy, precision, recall, F1-score and confusion matrix analysis. Experimental results show that the CNN model performs better, with a classification accuracy rate up to 97%, which is better than the SVM model. The trained model will be deployed using a web-based interface, where real-time prediction of diseases to the uploaded leaf images can be made.The suggested solution will offer a highly efficient and scalable means of early disease monitoring, lessening the need to employ experts and promoting accuracy in agricultural practices.