Explainable and Generalizable Multi-Crop Disease Detection Using Deep Learning and Self-Supervised Learning

The systematic review assesses the potential of deep learning and self-supervised learning to provide explainable and generalizable disease detection across different crops. Following PRISMA 2020 for abstracts, five databases which are Scopus, Web of Science,  and Google Scholar were searched from January 2015 to May 2026. The inclusion criteria encompassed studies involving image-based classification, detection, segmentation, domain adaptation/generalization, self-supervised pretraining, and explainable artificial intelligence for at least two different crops or transferable crop-disease settings. Studies involving solely classical machine learning techniques, non-visual symptoms, single-dataset toy experiments, inaccessible full texts, or no quantitative analysis were excluded.  A total of 2,846 records led to 2,118 distinct articles being screened, 214 papers undergoing full-text review, and 96 studies being ultimately selected, including approximately 3.42 million images, 42 datasets, 39 crop species, and 112 disease categories. Results were analysed both narratively and by outcome type.