Artificial Intelligence-driven Smart Postharvest Management of Horticultural Produce: Advances, Applications and Future Perspectives
Kumudani Sahu
Department of Fruit Science, Mahatma Gandhi Udyanikee Evam Vanikee Vishwavidyalaya, Durg Sankara-Patan, Durg - 491111, Chhattisgarh, India.
G. D. Sahu
Department of Fruit Science, College of Agriculture, Indira Gandhi Krishi Vishwavidyalaya, Raipur-492012, Chhattisgarh, India.
Mukesh Kumar Sahu *
Department of Floriculture and Landscaping, Mahatma Gandhi Udyanikee Evam Vanikee Vishwavidyalaya, Durg Sankara-Patan, Durg - 491111, Chhattisgarh, India.
Neha Sahu
Department of Soil Science, College of Agriculture, Indira Gandhi Krishi Vishwavidyalaya, Raipur-492012, Chhattisgarh, India.
*Author to whom correspondence should be addressed.
Abstract
Fruits and vegetables are highly perishable, and large quantities are lost after harvest through continued ripening, water loss, mechanical damage and microbial spoilage. Conventional post-harvest management depends on manual grading, visual inspection and fixed storage settings, which are subjective, slow and unable to detect internal defects or early spoilage. Artificial intelligence (AI), applied through machine learning and deep learning, has been studied as a way to make post-harvest operations faster, more consistent and non-destructive. This review examines recent advances in AI-driven post-harvest management of fruits and vegetables across four areas: quality assessment and grading; maturity, ripeness and defect or disease detection; shelf-life and spoilage prediction; and smart storage and cold chain management. Computer vision and hyperspectral imaging are used for grading and for predicting internal attributes such as soluble solids content and firmness; deep learning models detect surface and subsurface defects and classify ripeness; electronic noses combined with machine learning track spoilage and estimate remaining shelf life and the Internet of Things (IoT) combined with AI supports monitoring, prediction and traceability during storage and transport. Reported accuracies in individual studies are frequently above 90%, but several limitations restrict practical use: poor generalisation across cultivars, seasons and operating conditions; dependence on large labelled datasets; limited model interpretability; high equipment cost and the absence of shared datasets and common definitions, which makes published results difficult to compare. The review concludes that the contribution of AI to post-harvest management will depend less on further gains in laboratory accuracy and more on developing affordable, data-efficient and robust systems suited to real operating conditions, particularly for smallholders and developing countries where post-harvest losses are highest.
Keywords: Postharvest losses, machine learning, computer vision, hyperspectral imaging, shelf-life prediction, cold chain