Machine learning in agronomy: creating a web interface for early plant disease research

Abstract

A pressing issue in the digitalization of the agro-industrial complex is the fragmentation of Diptych tools, which prevent agronomists from quickly obtaining a comprehensive picture of the phytosanitary condition of crops based on comprehensive analysis data. The aim: to develop an architectural solution and create a functional prototype of a web platform to support decision-making in the early diagnosis of plant diseases by integrating heterogeneous data and machine learning model predictions. The study utilized systems analysis and object-oriented design methods. ASP.Net Core was chosen for backend development, while Next.JS was used for frontend development. The data streams from multispectral cameras, microclimate sensors, and neural network classification models were integrated. The main result is the developed modular microservice architecture of the future system and a prototype of the key interface, including a mapping module and a vegetation index visualization scheme. The full technology stack for implementation has been defined and validated. The approved platform concept lays the foundation for the creation of a tool that, once fully implemented, will improve monitoring efficiency by aggregating data in a single interface. Future work involves the phased development of services, integration with ML models, and field testing in a greenhouse complex to validate the approach.