AgriFoRwArdS Cohort 2 student Bradley Hurst, recently had the opportunity to visit Jersey as part of his industrial sponsorship with Jersey Farmers’ Union to progress his research in new active robot perception solutions for the automation of potato planting. In this blog post, Bradley shares some of the highlights of the trip.
During my latest trip to Jersey, I carried out data collection on Jersey Royal potato tuber under varying lighting conditions, in various storage configurations, and various points of view that the Jersey Royals are expected to be seen by the robotic manipulator to further develop the vision system. In addition, a data set was collected on the tracks the robot must be able to plant into in the field.
(L) The tubars stacked in the crates for planting ( and imaging). (R) Facilities and real life sceneries these systems need to work in on a day to day basis.
The trip was extremely useful for my research. These kinds of tasks can be extremely weather dependent, and due to the weather-dependent nature of the task, there would not have been a second opportunity to collect data of this quality and extent this year. Luckily, the weather turned out to be great, and by the end of the trip, I had been able to collect 10,027 RGB-D and NIR images of various potato tubers in different scenes, and around 5200 RGB images of tracks to plant the tuber into.
Whist visiting the island, I had the opportunity to get to know my industrial partners better, along with some of the local farmers who supported me in my data collection task, facilitating the requirements of my capture setup. This collaboration was essential in ensuring that the data collected were of high quality and framed in the correct context for the project.
This trip has significantly helped in carrying out my research. Up until now, I had an extremely limited amount of data to collect. Additionally, due to poor weather on my previous visit, a lot of the data was not collected under ideal circumstances, neither in the context of how the perception system being developed would see the tuber, nor under the correct lighting conditions. This time, the data capture was significantly more successful. The data captured is not only plentiful but carefully designed to frame the problem for the current project, which is important not only for the development of an instance segmentation approach but also for self-supervised learning, in which the design of the dataset is essential.
(L) The setup of the data collection process (R) The camera equipment used to capture the data.
One of the things I always enjoy while visiting Jersey is seeing the operations and talking to the experts that the work we are carrying out is being developed for. It’s an opportunity to discuss challenges being faced and get a different perspective on them. I also had the opportunity to get to know Millie and Lyra, two border collies that resided where I was staying while visiting.
(L + M) Lyra and Mille, the friendly border collies! (R) Jersey Shore
Overall, my trip to Jersey was a great success. The data collected will be essential for developing the vision system for the robotic manipulator, and the collaborations formed during the trip will provide valuable insights and knowledge for future work.
Bradley Hurst, AgriFoRwArdS CDT
A huge thank you to Bradley for taking the time to share his experiences so that other CDT students can also gain the benefit!
Do you want to get involved in the CDT?
If you are interested in learning more about what we do and if you share a passion for agriculture and technology then go to our AgriFoRwArdS CDT website to see more about our research, how you can be involved, and how to apply to be a student in the program.
Be First to Comment