Image based high-throughput phenotyping of crops using neural networks
Deep learning is a rapidly evolving technique with the ability to automatically extract features and analyse complex multi-dimensional datasets with high scalability to big data. It can be applied to a wide variety of plant phenotyping tasks, such as the identification of abiotic stress signs caused by climate fluctuations, helping improve crop monitoring and breeding aspects. Climate change has affected the yield of several major crops worldwide, for example as temperature rises, wheat yield is expected to reduce 5-6% per 1 °C of temperature increase. These environmental changes will cause significant abiotic stress to crops causing yield and quality decline, urging crop breeders to develop better adapted varieties in order to withstand the abiotic stresses. In this project, I will develop a model capable to identify abiotic stress in crops under field condition, using computer vision and deep learning.
The model will have application for crop breeders and growers. It will enable high-throughput phenotyping (HTP) of field multispectral images, collected by Unmaned Aerial Vehicles. The model will support crop breeders to automatically phenotype the varieties under trial, assisting the identification of climate-resilient varieties. Farmers can also potentially implement the model to monitor the crops development, allowing for better informed financial planning and overall management of the farm.
Develop a model capable to estimate the mean yield per variety based on multispectral image and weather data.