AI Based Approaches for Fruit Freshness Detection: A Systematic Review

Abstract

Different AI models are the latest technologies for detecting fruit freshness. They are a promising solution for reducing harvest losses, improving food quality assessment, and supporting smart agricultural practices. This study systematically reviewed 27 peer-reviewed articles and conference papers published between 2020 to 2025 on AI-based fruit freshness detection, using PRISMA methodology.  The review found that models such as ResNet, EfficientNet, and Transformer-based hybrid architectures achieved the highest classification accuracies, ranging from 97% to 99% under controlled laboratory conditions. The study also revealed that YOLO-based models generally demonstrated slightly lower accuracy but offered strong real-time performance, making them highly suitable for robotic and IoT-based agricultural applications. Furthermore, hybrid and multi-task learning frameworks were less frequently explored but showed significant potential due to their balance of accuracy, efficiency, and task diversity. This review also identified important limitations in existing studies. Controlled environments, homogeneous backgrounds, and stable lighting conditions are major characteristics of most datasets, which may limit the generalizability of these models in real-world scenarios such as farms, transportation systems, storage facilities, and marketplaces. Additionally, inconsistencies in evaluation metrics across studies make direct comparisons difficult. This review provides a comprehensive synthesis of recent advances and research gaps in AI-based fruit freshness detection. The findings suggest that the effectiveness of AI-based fruit freshness detection models depends on the application context and deployment requirements. Future research should focus on developing more diverse real-world datasets and adopting standardized evaluation protocols to improve the reliability and practical applicability of these systems.