NIVA Geotagged Photos AI Platform and Irish Challenge (privacy and pasture identification)

As part of the NIVA project, a platform for AI powered auto review and pre-validation of geo-tagged agriculture images for IACS (Integrated Administration and Control System) will be developed. This platform, described in this presentation,is developed by TheWalton Institute/ South East Technological University,on behalf of NIVA. The platform is split into two main aspects, a general facility for image uploading and tagging, and a facility for reviewing by a human or auto-reviewing by an AI model. For the targeted initial three auto-review (AI) tasks, suitable auto-review challenges have been proposed by Paying Agencies. Including the Irish challenge (pasture identification)for which the model training and validation is described in this document. In Ireland it is already possible for farmers to use images from a geo-tagged app to prove presence of pasture. However, this is resulting in many images being sent in and needing manual inspection for approval. A possible pre-selection by an AI model is helpful. This document summarises AI work on pasture identification and soil quality. Additionally models trained to detect and obscure faces (privacy preservation) are described. All source code is available in NIVA gitlab -–https://gitlab.com/niva-ai-platform. A full wiki is also available -–https://gitlab.com/niva-ai-platform/infrastructure/-/wikis/Architecture.