Advances in Machine Learning and Deep Learning Techniques for Grape Disease Classification: A Comprehensive Survey

Abstract

The health and productivity of a vineyard can be maintained through the early diagnosis of grape disease. Incorporating machine learning (ML) and deep learning (DL) methods has demonstrated to be a very effective strategy for this challenge. The recent developments in DL and traditional ML techniques for the recognition and categorization of the grapevine disease are discussed in this review. Alongside the traditional ML methods like fuzzy logic systems and Support Vector Machines (SVMs), a variety of methods have been recognized, which include Convolutional Neural Networks (CNNs), Generative Adversarial Networks (GANs), capsule networks, residual networks and hybrid methods. Hyperparameters have been adjusted through optimization algorithms which include Improved Salp Swarm Optimization (ISSA), while the improvement of model efficiency for real-time and mobile applications is finished by attention mechanisms and lightweight architectures. Furthermore, GAN-based data augmentation methods resolve the limitations of small datasets. Even with advancements in robustness and classification accuracy, problems regarding computational efficiency, generalizability, and real-time deployment in natural environments still exist. The strengths and weaknesses of these techniques are thoroughly analyzed. In this review, providing awareness into future research for grape disease detection.