Intelligent transformation of mango maturity Estimation: AI-based non-destructive methods

• A comparative analysis of NIRS, HIS, image processing, and DL techniques for estimating mango maturity. • Distribution analysis showed that NIRS accounted for 61.5% of the studies. • AI enables accurate prediction of mango maturity and enhances spectral data calibration. • A roadmap toward integrated, scalable, and sustainable mango maturity systems. Estimating mango maturity is critical for determining the optimal harvest time, controlling post-harvest operations, and meeting consumer demand in the market. However, non-destructive techniques often provide limited accuracy in maturity estimation. This accuracy can be improved by integrating these methods with artificial intelligence models. This review systematically evaluates these methods based on the PRISMA-ScR guidelines in terms of Near-Infrared Spectroscopy (NIRS), Hyperspectral Imaging (HSI), image processing coupled with Machine Learning (ML), and Deep Learning (DL). Distribution analysis showed that NIRS technology accounted for 61.5% of all studies, followed by image processing at 16.7%, DL at 12.5%, and HSI at 7.3%. NIRS and HSI calibration are still prevalent, utilizing statistical and chemometric models. Over the last few years, the integration of DL with NIRS and HSI has enhanced preprocessing and calibration efficiency, and image-based DL models have been developed as strong predictors of mango maturity. Nevertheless, relatively little research has been conducted on attention-based DL models, indicating that these models are still in the exploratory stage. Comparative analysis reveals that imaging and DL techniques are highly accurate in estimating maturity for some varieties, while NIRS and HSI techniques are still superior in predicting internal quality (total soluble solids, dry matter, titratable acidity). Despite these advancements, limitations remain regarding cost, calibration transfer, and ease of field deployment. To make mango maturity estimation systems robust, scalable, and sustainable, future research should prioritize the integration of multi-sensory data, the application of lightweight artificial intelligence models, and the exploration of alternative approaches to overcome the limitations of current methods.