Analysis of crop images using deep learning (Industry Project with Muddy Boots Software)

Nouf Alkahtani


Supervised by Padraig Corcoran; Moderated by Xianfang Sun

This project will be performed in collaboration with the Muddy Boots Software company. The following is the project description - Through our Grower Management platform, we have a large dataset of images taken in the field displaying weeds/pests/diseases in crops, which are tagged against the specific problem with the severity. We would like to run a project to see if we can create a Machine Learning model that can detect a weed/pest/disease based on an image taken on a device - this is to help users who may not have the agronomic knowledge to identify what problems they have in their crop. We have already run a Proof of Concept with Python + Tensorflow to validate that there is a valuable use case and we have suitable data. The data of thousands of images needs to be categorised by crop and problem, then cleaned to make sure we do not have false images skewing the model. Our model does not take crop into account and is only using top 6 problems - we'd like this to cover a wider range of crops and problems. The model would run every 1 or 2 weeks in production, but we would not expect a production-ready system as a result of this project. It would be to help us validate the data and technology, so we can demonstrate this back to potential customers. The Muddy Boots company HQ is based in Ross-on-Wye, and they would prefer that the student was based here, but could also run remotely or at Bristol office if needed. A non-disclosure agreement (NDA) may need to be signed by the student completing the project.

Final Report (23/10/2020) [Zip Archive]

Publication Form