An Iterative Systematic Analytical Review of Advanced Deep Learning Models for Grape Disease Detection

Abstract

The growth in plant disease occurrence has heavily threatened global agricultural productivity. Therefore, highly efficient and automated diagnostic tools need to be developed in the current scenario. Rapid advancement in machine learning and deep learning techniques makes present reviews far from conducting comprehensive comparative evaluations of cutting-edge models especially for important crops such as grapes related to plant disease detection. These reviews inadequately address the challenges of model interpretability, computational efficiency, and adaptability to various datasets, thus opening the doors for gaps in applying such technologies in actual farm-related settings. This paper seeks to address these limitations by doing a -guided systematic review of the latest ML and DL models including CNNs, Vision Transformers (ViTs), and hybrid ensemble approaches like DenseNet121, MobileNetV3Large, and Inception-ResNet-V2. The performance is evaluated in terms of major metrics such as accuracy, precision, recall, F1-score, and computational efficiency. Some of the new architectures include dual-track networks with Swin Transformers and group shuffle residual deformable nets, among others like interpretable models LEViT and an interpretable leaf disease detector, I-LDD, which are specifically discussed in terms of new architecture and practicality benefit. In the light of impact on scalability and usability in agricultural domains, the emphases go to lightweight architectures that optimize deployment for edges and explainable frameworks improve decision-making capabilities. With this, an inclusive review proves helpful for better selections of optimum models toward the detection of grape disease as well as generalized plant pathology with precision advancements in agriculture scenarios. This work contributes to sustainable agricultural practices and enhances food security by bridging knowledge gaps and proposing scalable, efficient, and interpretable solutions.