A comprehensive system supporting sustainable agricultural production from farm to fork

Advancements in Artificial Intelligence (AI) and Computer Vision resulted in significant applications in precision agriculture for high-value crops. However, the full potential of AI is still untapped, particularly in regions with low-income agricultural production, where high-end systems for systematic surveying and treatment are often unaffordable. Our ongoing research focuses on developing accessible computer vision methods that substantially impact the farm-to-fork strategy in view of the United Nations Sustainable Development Goals, especially for “Zero Hunger” [1].
Even though image classification systems have achieved accuracy levels that often surpass human performance – as demonstrated in our previous work [2] – there remain significant challenges and room for improvement. Nonetheless, in the long run, a shift from basic AI-assisted monitoring to more advanced approaches that deliver deeper insights and actionable knowledge is required both from a scientific and applicative standpoint. Can we foresee changes in state after identifying and classifying an object? Can we accurately predict how many plants will reach flowering or fruiting within a specific period, allowing us to match production with demand better, optimise revenue, and minimise food waste? These tasks become particularly complex when dealing with crops planted in open fields and greenhouses located remotely and experiencing varying light, weather, and risk conditions. 
An integrated set of methodologies is essential for achieving these goals. First, we need robust techniques to detect and classify plants, distinguishing desired crops from weeds. We also require ongoing monitoring to detect early signs of disease, nutrient deficiencies, stress, or growth anomalies and to take preventive measures when possible. Lastly, tracking and forecasting plant’s flowering and fruit ripening stages is needed to optimise harvest timing and yield.
By combining advanced computer vision, mobile computing, remote sensing, and predictive analytics, we aim to create a comprehensive system supporting agricultural production in diverse settings, addressing immediate and long-term challenges.

[1] Leadership Council of the Sustainable Development Solutions Network, “Indicators and a Monitoring Framework for the Sustainable Development Goals,” 2015, available
at https://sustainabledevelopment.un.org/content/documents/2013150612-FINAL-SDSN-Indicator-Report1.pdf.  Last retrieved April 29, 2024.
[2] A. Bruno, D. Moroni, R. Dainelli, L. Rocchi, S. Morelli, E. Ferrari, P. Toscano, and M. Martinelli, “Improving plant disease classification by adaptive minimal ensembling,” Frontiers in Artificial Intelligence, vol. 5, p.
868926, 2022.