The Miscanthus AI project (a research partnership between Aberystwyth, Lincoln and Southampton universities) aims to assess whether artificial intelligence can help in plant breeding. This ranges from simply automatically measuring crops in the field to mimicking the decisions of a highly skilled plant breeder.
With rapid progress in sensor technology, scientists are looking into phenomics as a way of predicting future generations of plants much quicker than traditional methods. This could even replace genomics because instead of using DNA, the technology measures specific plant characteristics. This includes near infra-red imaging which can predict the chemical and genetic composition of plants in a less invasive way than previous techniques.
With the threat of climate change and the challenges that lie ahead for farmers, accelerating the plant breeding process is key to improving crop performance in line with changing environmental conditions. Using AI predictive technology will play a major role in this. As markets change, creating new resilient varieties rapidly goes some way to improving food security.
What equipment is used within the Miscanthus AI project?
Before developing machine learning for decision-making tasks in plant breeding, the crops in question first need to be analysed and measured. For Miscanthus, the height of the crop determines the above ground yield which is what the farmer is looking for when growing a biomass crop.
Measuring Miscanthus was traditionally done using a very long ruler as it can grow up to 4 metres every year. This is highly laborious and time intensive with the potential for a great amount of error owing to the inability to see the top of the plant from ground level.
In the Miscanthus AI project, drones are being used to measure the crop growth. The drones are flown above the field with a sensor attached which measures the heights of the Miscanthus.
The Kit
DJI Matrice 300 RTK drone
Zenmuse L1 LiDAR sensor
MicaSense RedEdge camera
A pre-determined flight path is plotted using a map interface. This ensures that the drone follows the exact same route every time it is flown over the plots and means that the different images created at different time points are 100% comparable to one another.
The drone can only carry either the sensor or the camera at any one time. As it flies over the Miscanthus with the LiDAR sensor at a height of 15 metres, the laser is fired downwards and the time it takes for the laser to return to the sensor determines the plant’s height. These data points create what is called a point cloud. The Real-Time Kinematic (RTK) aspect of the drone gives a really accurate GPS reading, plotting a position within a few centimetres. The camera takes pictures every few metres as it flies over the plots.
Outputs created
The output created is a full colour 3D model of the vegetation. All the data points and photos are stitched together, with the images stuck on top of the point cloud to create a more accurate 3D representation.
Every time the drone is flown to gather data (once per month), truth data is also collected for comparison. Ten plants are cut, their length measured, leaf colour and number recorded, and biomass calculated to validate the process.
Using drones to gather all of this information is the first step to automating the plant breeding process. It is hoped that future applications will include a computer-generated “breeder’s eye” to select parent plants for the next generation.
The data collected through the Miscanthus AI project follows the plants from emergence all the way through to senescence. This comprehensive data set will be made publicly available at the end of the project and enhances our understanding of Miscanthus biomass. It is also hoped that the project will provide the basis for exploring what other phenotypic traits can be extracted from the data.
Other areas drone technology is used in plant science
Researchers at Aberystwyth University are collaborating with those at Scion in New Zealand. Scion have developed a deep learning tool for use in forestry and beyond. They have called it Forest Insights. This involves using LiDAR and satellite imagery to detect the species of trees in commercial forests, their age, size and health. This automates what is otherwise a
labour-intensive process used in mapping forests. It helps companies manage their inventories and to make strategic decisions for the timber supply chain.
Scion is also using the technology to identify potential plant diseases in New Zealand. Myrtle rust threatens myrtle trees, one particularly precious species being the iconic pōhutukawa. The rust can spread easily if not acted on quickly. Scion’s technology can not only identify the myrtle trees and those at risk, but the remote sensing technology detects infected trees a day before physical symptoms is visible to the naked eye. For example, heat loss from damaged leaves can be picked up on thermal imaging cameras.
Bringing all this technology together and using drones to fly over large swathes of woodland is key to improving the forestry industry in New Zealand and, hopefully, in other parts of the world.