A PRISMA-guided systematic review of artificial intelligence techniques for fruit diseases detection

Abstract

Fruit diseases pose a serious challenge to agricultural productivity and require effective and prompt detection strategies. Traditional techniques based on manual observation are often limited by subjectivity and delay, creating a strong need for automated solutions. Deep learning and artificial intelligence (AI) have become revolutionary technologies in detecting fruit diseases, which can be correctly classified and diagnosed at an early stage in various agricultural environments. This study conducts a systematic literature review guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) to summarize the recent achievements in the field. Unlike prior reviews focused on fruit grading, this work centers on pathological diagnosis. Our initial search across Scopus and Web of Science databases yielded 876 records. After applying inclusion and exclusion criteria, 31 primary studies published between January 2024 and August 2025 were selected. These papers were further analyzed to obtain three core areas, namely recent methods, ongoing challenges, and future research directions. Recent methods demonstrated strong progress with attention mechanisms, transformer hybrids, and YOLO-based detectors specifically optimized for real-time use on mobile phones and drones. However, practical application is heavily hindered by data scarcity and poor cross-domain generalization. In addition, moving these systems from the lab to the farm is often restricted by hardware limitations and a lack of explainable AI. Thus, future research must prioritize diverse, multi-site datasets and ensure that model outputs make practical sense to agronomists. This review consolidates these findings to help researchers build scalable, interpretable, and field-ready disease detection tools.