Deep Learning in Precision Phytopathology: A Comprehensive Survey of CNN Architectures for Disease Detection and Severity Quantification

Plant diseases are a major threat to the global agricultural productivity causing considerable yield losses and economic damage. Recent developments in Artificial Intelligence (AI) and specifically in Deep Learning has led to a revolution in the diagnosis of plant diseases with automated and scalable analysis of the crop images. This review offers an extensive synthesis of Convolution Neural Network (CNN) architectures engineered in precision phytopathology and mainly aimed at disease detection, classification, localization, and severity quantification. Following a systematic literature review using PRISMA methodology, this paper reports a structured taxonomy of CNN-based methods including classical classification networks, object detection networks and semantic segmentation models from 131 peer-reviewed articles. The review compares popular architectures like ResNet, EfficientNet, Yolo and U-net, highlighting their performance characteristics, accuracy-efficiency trade-offs, as well as suitability to real-world deployment. Findings show that although CNN based systems hold a high diagnostic accuracy in controlled settings, there are still issues especially related to generalization, dataset availability, and field level robustness. Emerging directions such as hybrid CNN-Transformer models, multimodal sensing and edge deployment are found as critical enablers for next-generation precision phytopathology systems for sustainable and data-driven crop disease management.