Automated platform for detecting and mapping crop diseases using UAV and Artificial Intelligence

With the increase of the world’s population, the constant growth in food demand generates the need to find new ways of optimizing agricultural workflows. Today, a wide variety of software is dedicated to precision agriculture that helps the farmers gather data otherwise hardly available and extract information to minimize crop losses or prevent diseases. However, these solutions hardly allow a complete workflow from gathering the images, processing the datasets, and mapping and detecting the crop diseases. The solutions can be precisely applied where needed, without the need of manually exchanging information between applications. In this paper, we are proposing an architecture as well as a possible implementation for a web platform that can manage such workflows. The platform was implemented using ASP.NET Core 3.1 with C# as the main programming language. Following the best practices in terms of maintainability, the integration with third-party software was developed using proxy components that implement each components’ SDK or API, making these external solutions easily interchangeable. The first module presented in the paper covers the integration of third-party UAV controlling platforms. We integrated the UgCS commercial solution using the provided .NET SDK for our scenario. The data gathered by the UAVs controlled by this module, consisting of RGB, thermal and multispectral images, were stored using Azure Blob Storage cloud service. The location of each image data was acquired by extracting the XMP metadata and further stored using a PostgreSQL database with the PostGIS extension. Next, we provided a way to automatically generate the orthophoto imagery from the acquired data by integrating the Python API available in Agisoft Metashape. Lastly, using the products obtained from the previous step, we calculated different vegetation indices of the analyzed fields using C# and analyzed the outcomes using deep-learning models to identify and map vegetation health states. The platform has been implemented and tested for several case studies located across Romania, reaching satisfactory results.