A Systematic Review On Deep Learning Techniques For Maturity Classification of Tomato Fruit On Plants

With the development of smart agriculture, automatic on-plant maturity grading of tomatoes has become a key factor in optimizing harvesting decisions. Based on the PRISMA 2020 guidelines, this study systematically reviews 42 high-quality papers published before September 10, 2025, aiming to address the fragmentation of existing research in terms of methodology, datasets, and evaluation criteria. The study found that: (1) the technical path has achieved a leap from manual features to end-to-end deep learning (such as R-CNN, YOLO, and Transformer), significantly improving detection performance in complex environments; (2) datasets are still mainly self-built, generally facing challenges such as inconsistent annotation standards and a lack of scene diversity; (3) evaluation criteria have shifted from a single accuracy indicator to a multi-dimensional balance of "accuracy-power consumption-latency". The conclusion indicates that lightweight model design and edge deployment are the core of realizing the transformation of laboratory technology to the field. Future research should focus on the construction of large-scale multimodal datasets, robustness enhancement in complex backgrounds, and multi-task collaboration in harvesting decisions. This study provides a systematic technical reference for the intelligent upgrading of the tomato industry.