Research project Intelligent Spraying with explainable AI for precision farming
General introduction
The POSING research project has produced a series of validated AI models for powerful real-time weed detection and intelligent spraying. The models are supplemented with Explainable AI that explains choices and integrates farmers' knowledge. This opens up prospects for more reliable precision spraying in the field, i.e., only spraying weeds where necessary. The result is a reduction in environmental impact and increased efficiency in resource use. The project, part of the Flemish AI research program (FAIR), aimed to make AI innovations, such as Vision and Explainable AI, practical and accessible to companies within six months. In collaboration with market leaders such as Delvano and Spray Venture, the project also focused on practical applications, particularly weed detection.
Research approach
High-resolution cameras, drones, and sensors were used for image recognition to identify weeds in real time and send spraying instructions. Explainable AI provided textual explanations for decisions and processed the farmer's tacit knowledge. The project results were validated through field trials, performance comparisons with existing solutions, a feedback loop with the involved farms, and extensive data analysis. These methods ensured the robustness, accuracy, and practical applicability of the AI models for weed detection and intelligent spraying.
The project results were validated through real-world field trials, performance comparisons with existing solutions, a feedback loop with participating companies, and extensive data analysis. These methods ensured the robustness, accuracy, and practical applicability of the AI-models for weed detection and intelligent spraying.
Relevance/Valorization
The need for Vision-AI in intelligent spraying technology extends beyond the partners involved: farmers want to optimize cultivation with minimal environmental impact. Explanations of system actions and user feedback increase confidence and adoption. The weed detection and spraying results developed serve as a benchmark and case study with concrete lessons for practical AI integration in agriculture.
