Paper Details

Abstract

Wheat is one of the most important staple foods in the world, supplying more than one seventh of the calories for global human consumption. However, wheat production is routinely affected by foliar diseases including leaf rust, stripe rust, powdery mildew, and so on. In addition to reducing crop quality, such diseases cause significant economic losses if they are not timely found and controlled. Conventional detection techniques such as manual field survey or laboratory diagnosis are usually labour-intensive, subjective and inapplicable for large scale agricultural surveillance. Thus, to mitigate these obstacles, deep learning (DL) techniques have been proposed as possible solutions for automated, accurate, and scalable progression using image-based analysis. In this review, we systematically summarize recent development of deep learning in wheat leaf disease detection. It covers common model architectures including Convolutional Neural Networks (CNNs), EfficientNet, and Vision Transformers (ViT) and involves comparison to feature extractor, evaluation metrics (accuracy, precision, recall, F1-score, IoU), and deployment issues are also discussed. Despite encouraging accuracies in artificial environments, there still exist several issues, including limited generalization of the models in real environments, lack of interpretability and absence of standardised benchmarks. In addition, this work can help researchers and practitioners to develop reliable interpretatable and field-ready deep learning based solutions that boost sustainable wheat farming.

To Cite This Article