DeepDetect: Welsh research explores smartphone AI for earlier potato blight detection

By Lukie Pieterse, Potato News Today

Researchers at Aberystwyth University in Wales have investigated how artificial intelligence and images of potato leaves could help growers recognise disease earlier, through a project designed around the practical needs of farmers and agronomists.

Called DeepDetect, the initiative brought together computer science and biological research to assess the feasibility of a mobile diagnostic tool for potato crops. The funded project ran from June 16, 2025, to January 31, 2026, and is now listed as finished.

Its ambition was to make disease recognition more accessible, helping growers obtain useful information directly through their smartphones. Publicly available information, however, does not yet establish a commercial launch or demonstrate the system’s performance under field conditions.

Building a tool around the grower

The proposed system centred on machine learning trained using images of healthy and diseased potato leaves. Announced in July 2025, the project included plans to develop a prototype and refine it through workshops and focus groups with growers and agronomists.

Testing on Welsh potato plants and real samples was also part of the planned evaluation. A minimum viable app was to be presented to farmers and advisers for feedback, placing usability alongside biological and technical performance.

Project lead Dr Edore Akpokodje, a lecturer in computer science, emphasised the importance of involving the intended users from the beginning.

“By integrating farmer feedback from the outset, we will ensure that this technology is grounded in real-world needs and challenges,” he said.

That approach addresses a practical question facing agricultural technology developers: whether a promising research concept can become a dependable tool that fits the working day of a farmer or crop adviser.

Why earlier recognition matters

Late blight is caused by Phytophthora infestans, an oomycete, or fungus-like organism, capable of infecting potato foliage, stems and tubers. The pathogen can develop rapidly after infection, with visible symptoms appearing within days. Infected tubers entering storage can also lead to substantial losses.

That combination makes timely recognition valuable throughout the production chain. Disease forecasting, resistant varieties, control of infection sources and protection against tuber blight all contribute to integrated management.

An effective image-based diagnostic tool could potentially add another source of information to that process, helping growers identify suspicious symptoms and seek appropriate advice sooner.

The practical opportunity is to strengthen the information available to growers and agronomists when they assess crop health and decide how to respond.

Field performance remains the key test

The wider research challenge is translating image recognition into dependable performance under farming conditions.

Independent research on potato disease detection in Spain has highlighted the limitations of models developed using individual leaves photographed in laboratory settings. Field images present a more complicated task: foliage overlaps, symptoms may be small, and camera distance and surrounding conditions vary.

Those challenges help explain why testing with real plants and representative field images matters. A model’s ability to recognise disease in a carefully prepared image does not, by itself, establish how reliably it will perform during routine crop inspections.

For DeepDetect, the public information reviewed does not provide diagnostic accuracy figures, completed field-validation results or a confirmed commercial release.

The project’s original ambitions included detecting disease before signs became visible to the human eye. The available material does not establish that this capability was demonstrated.

Potential beyond a single crop

The researchers envisaged the feasibility work contributing to a broader potato blight warning system, with possible applications in other crops and regions.

For the potato industry, the project illustrates continuing interest in bringing disease assessment closer to the grower through widely available mobile technology.

Its eventual value will depend on evidence that the technology delivers reliable, useful information when crop-protection decisions must be made. Farmer participation can help shape that usefulness, while biological validation and field testing must establish what the system can consistently detect.