Challenge

Although plant-parasitic nematodes may be unfamiliar outside agricultural contexts, they are widely recognised within the sector as a significant source of crop damage and economic loss.
Plant-parasitic nematodes (PPN) are associated with approximately 5–20% yield loss across agricultural crops worldwide, contributing to an estimated US$175 billion in crop losses each year. When impacts on non-commercial crops in developing regions are included, the global cost is thought to rise to around US$200 billion annually. The challenge is particularly acute in the UK potato sector, where potato cyst nematode (PCN) continues to cause significant economic and production losses.

• England and Wales: Approximately 65% of land used for ware potato production is infested with PCN, contributing to an estimated £50 million in direct annual losses to the UK potato industry.
• Scotland: PCN infestation leads to an estimated opportunity loss of £5,000 per hectare, equivalent to around £25 million in annual losses to Scottish potato production.
• Projected impact in Scotland: Without effective intervention, annual losses are projected to rise to approximately £125 million by 2040.
• Scottish ware potato land affected: More recent analysis suggests that around 41% of Scottish ware potato land is already affected by PCN.
• Rate of spread in Scotland: PCN-infested land is reported to be doubling roughly every seven years.

These figures underline the scale and urgency of the problem. Improving diagnostics, non-chemical control strategies, and precision, data-driven pest management could reduce losses for growers, strengthen the resilience of food production systems, and support long-term food security. This is the challenge that UK-based company SparkSoft is working to address.

Solution

The SparkSoft team is creating an AI-driven platform that brings together image detection, predictive technology, and user-friendly interface to give growers practical and automated guidance on how to identify, quantify and forecast pests infections and emerging outbreak risks, enabling targeted interventions that significantly reduce unnecessary pesticide use and support more sustainable, data-led farming practices. To help push this project forward, SparkSoft founder Peng Yue joined Innovate UK’s BridgeAI programme, were Peng was matched with an Independent Scientific Advisor, Professor Po Yang, and a technology-focused intern, computer science PhD student Zhipeng Yuan, both from the University of Sheffield.

Across a three-month placement, Zhipeng worked on two interlinked parts of SparkSoft’s pest management platform, known as PPNAnalyzer, addressing a clear technical need. First, Zhipeng strengthened the data and computer vision modelling foundations underpinning the system, this included expanding and curating SparkSoft’s training dataset to ~5,000+ labelled images across six common plant-parasitic nematode (PPN) species, standardising annotation formats, improving class balance, and introducing clearer dataset splits and versioning to support repeatable training and evaluation. These updates improved overall dataset reliability (e.g., reduced duplication and inconsistent labels) and made the data easier for engineers and collaborators to maintain and extend. Building on the improved data pipeline, Zhipeng optimised the detection workflow to increase robustness and accuracy, reaching a 96% mean average precision, providing a stronger technical baseline for deployment and ongoing improvement.

Risk prediction and forecasting

Alongside image-based detection, Zhipeng also developed a prototype for PPN distribution prediction, translating environmental and soil covariates into spatial risk estimates. Using a maximum entropy (MaxEnt) approach, he integrated available layers such as climate variables and soil properties to generate probability surfaces indicating where each nematode species is most likely to occur. This work demonstrated how the platform can move beyond “what is in this sample?” toward “where is the risk likely to increase next?”, supporting more proactive pest management and field planning.

The result

“From SparkSoft’s perspective,the internship has been highly valuable for us. It provided targeted technical support at an important stage of our development and helped us strengthen the foundations of our AI-driven precision agriculture platform. One of the clearest benefits was support in developing our pest quantification and predictive modelling capability has significantly expanded our opportunities to deliver more effective and forward-looking pest management solutions”. says Peng.

“The intern also contributed to the development of a functional minimum viable product (MVP) of our user-facing AI assistant, this assistant is designed to help growers interpret complex pest-related information through natural language queries, for example, questions about pest species, infestation risk, or the influence of weather conditions.” Rather than relying on a single model, the system uses a multi-agent architecture, with dedicated components responsible for tasks such as identifying nematodes from images, predicting their likely spread, and searching relevant contextual documents. These components are integrated using a system known as retrieval-augmented generation (RAG) framework, enabling the assistant to ground its responses in verified data sources rather than producing generic outputs.

“This tool has given us a real sense of how language model features could be integrated into our products and helped us understand where these tools could benefit our stakeholders in agriculture. Zhipeng’s clear academic documentation and experiment logs were also particularly valuable for us as a small company, as they improved visibility of research progress and supported more informed decisions around technical focus and resource allocation.”

Wider impact and next steps

Through the internship, Zhipeng strengthened his technical skills in areas such as multi-agent systems and the application of large language models, particularly in a business-focused context. This, he says, will be of great value for his longer-term career development, in which he aims to combine academic research with solving practical industry problems.

Po, who has a longstanding relationship with SparkSoft, played a key role in helping Zhipeng understand how academic AI models can work in industry settings. He adds: “The internship progressed very well, and Zhipeng quickly became an effective and proactive member of SparkSoft’s research and development team. He made a significant contribution by enhancing the company’s nematode recognition and analysis platform, integrating state-of-the-art AI techniques combining pattern recognition with rule-based reasoning. This has improved the functionality, transparency and auditability of SparkSoft’s platform. Zhipeng has also engaged effectively with external research partners in the sector such as ADAS and Fera, helping to align the technical work with real-world agricultural and regulatory requirements.”

As a direct outcome of Zhipeng’s work during the internship, SparkSoft secured a three-month Innovate UK Sovereign AI proof-of-concept award to further develop its platform. The Sovereign AI initiative aims to validate scalable AI technologies with the potential to make an impact in areas of strategic importance to the UK.