Automated sensing and monitoring of agricultural environments has the potential to improve farming practices and increase the yield and sustainability of food growing processes. Currently, most agricultural automation relies on robotic platforms such as ground vehicles or aerial drones. While these have shown a great promise for fine-grained monitoring of agricultural surfaces, they cannot be deployed autonomously under the current regulatory framework. Data collection is instead typically performed by a certified operator, and even systems capable of autonomous navigation and data collection require constant monitoring. Additionally, imaging data collected from different robotic platforms are usually taken from different viewpoints and using different imaging sensors, making it difficult to design a system that can integrate these different sources of data.
Satellite images have been widely used for monitoring agricultural areas, and can highlight areas in need of closer attention and provide prior information to more fine-grained monitoring or intervention solutions. While the volume of available earth observation data has steadily increased following improvements to imaging technology, there is still an unmet need for remote sensing approaches that can effectively utilise, interpret, and make quantitative decisions based on this data. In this project, you will investigate the feasibility of hyperspectral imaging to remotely estimate parameters (potassium, phosphorus pentoxide, magnesium, pH) of agricultural soils. Unlike conventional photographs with three colour channels (red, green & blue), hyperspectral images consist of many channels where each corresponds to a narrow wavelength band in the visible/near-infrared electromagnetic spectrum (400 – 1100 nm). You will compare different approaches for determining these different parameters and establish a robust baseline which will serve as a basis for future research endeavours, including integration with data collected by autonomous robots at ground level. Prior knowledge of Python and image processing libraries (such as OpenCV) is desirable for this project.
The project would be supervised by Dr Petra Bosilj and Dr James Brown, both of whom have prior experience applying computer vision and machine learning approaches to numerous application domains, ranging from medical imaging, precision agriculture, phenotyping and ecological conservation. For more information, please reach out to Petra via e-mail (firstname.lastname@example.org) or Teams.