A commercial glasshouse trial reported a 69 per cent alignment between AI-generated sensor alerts and grower observations of spider-mite presence, marking the most detailed validation to date of Altered Carbon’s machine-learning platform for early pest and stress detection in tomatoes. The result comes from the completion of the TomatoGuard project, an Innovate UK-funded initiative led by Altered Carbon with APS Produce, Fargro and the UK Agri-Tech Centre.
The programme combined graphene-based volatile organic compound sensors with embedded electronics and machine-learning models, integrated into a cloud system designed to interpret plant-emitted VOC patterns as indicators of stress or pest activity. Trials in laboratory, controlled-environment and commercial settings produced data that was used to train and evaluate the platform.
In commercial deployment at APS Produce, early-stage alerts were influenced by connectivity resilience, battery autonomy, sensor positioning and crop canopy ventilation, all of which were identified as factors affecting data quality and operational stability. Altered Carbon said the trials enabled a transition from prototype hardware to real-crop testing and provided information on how glasshouse conditions shape model performance. Planned development areas include revised hardware, alternative connectivity options such as LoRaWAN and 4G, and expansion of the training library to cover additional pests and stress signatures.
Fargro independently monitored the commercial trial and reported that the sensors showed potential to detect spider mite, adding that improvements to battery life and Wi-Fi stability would support broader testing of the system’s capacity for pre-symptomatic detection. APS Produce used the Scent Studio platform to review alerts and sensor behaviour and stated that automated detection tools could support crop monitoring where workforce turnover affects manual scouting.
The UK Agri-Tech Centre, which conducted controlled-environment studies to generate training and validation data, said the approach to monitoring plant signalling may provide a route to reducing pesticide and nutrient inputs. The project generated new datasets, hardware and software assets and established partnerships intended to support further development and possible expansion to other crops and disease indicators.
