Harnessing hyperspectral and thermal camera data to detect crop disease early

The outcome

Pioneering algorithms and methodologies developed at the Hyperspectral and Thermal Remote Sensing Laboratory (HyperSens) will detect critical changes in vegetation by interpreting the complex data produced through hyperspectral and thermal imaging. These technologies capture changes in the physiology of the plants long before they can be detected using pre-existing tools, with hyperspectral imaging detecting information from the electromagnetic spectrum invisible to the human eye.

HyperSens Lab is established jointly between the University of Melbourne’s School of Agriculture, Food and Ecosystem Sciences in the Faculty of Science, the Department of Infrastructure Engineering and the Faculty of Engineering and Information Technology (FEIT).

Head of HyperSens Lab Professor Pablo Zarco-Tejada explained that the technology will have a broad application across both crops and forestry.

“The hyperspectral and thermal images detect changes in transpiration rates and in photosynthetic pigments that absorb light. We can tell that a plant – an olive tree, an almond tree, or other species – is infected months before a plant pathologist or any technician in the field can visually detect that something is going wrong.”

The new technology can be used for both identifying diseases as well as abiotic-induced stress, such as a lack of nutrients or water. And crucially, the algorithms and methodologies being developed by Professor Zarco-Tejada’s team can pinpoint the issue where previously, symptoms could be easily confused and mistreated.

“There are symptoms that look like a particular disease but could be due to bacteria, a virus, or a lack of water or nutrient. With our algorithms, we can distinguish the issue – and we can use this data to provide advice to growers or those trying to use water more efficiently.”

Learn more about Food and Agribusiness research

The need

With global warming and increased traffic between countries over the last 50 years, the number of plant pathogens and outbreaks is increasing.

The number one threat to Australian agriculture is a bacteria called Xylella fastidiosa (Xf) – which produces a currently incurable disease that causes plants to wither and possibly die. Xf has devastated the olive plantations in Apulia, Italy, and it is currently affecting other countries in Europe such as Spain, France and Portugal. If it were to spread through Europe and to the Mediterranean basin, the losses in just the olive industry alone are projected to reach up to €5.2 billion per year.

Early detection is necessary to contain Xf, but some infections don’t cause visual symptoms until 8-10 months after infection – even though the plants are infectious during this time.

However, the algorithms developed by HyperSens lab are taking us one step closer to accurate, large-scale screening of Xf and other harmful diseases by detecting the physiological changes that start happening once plants are infected.

The research

Much of the research to date on hyperspectral and thermal imaging has been focused on the development of the devices, including hyperspectral imagers and thermal cameras. At the University of Melbourne, Professor Zarco-Tejada and his team operate a sub-nanometer hyperspectral imager, which collects data in very narrow spectral bands measuring a 0.1 nanometer width – 100 times thinner the spectral bands achieved by any standard hyperspectral sensor.

Now, HyperSens Lab is turning its attention to developing algorithms and methods that interpret the data, as so much is collected that it becomes difficult to handle properly.

“We are evaluating and investigating which spectral regions and bands are linked to the physiological changes that are happening under infection. And then we can develop and fine tune the models to be able to work properly and to be used operationally.”

In one project, Professor Zarco-Tejada is supervising PhD students who are developing algorithms to quantify chlorophyll fluorescence accurately in hyperspectral images. His team have shown in their research publications that detecting these signals is extremely important when trying to identify the stress of plants at earlier stages of stress due to disease or lack of water.

In a partnership with the National Institute for Forest Products Innovation, HyperSens Lab are also researching the role of chlorophyll fluorescence signals in monitoring and mapping variability of nutrients in forest vegetation. These methods could potentially be used to more accurately fertilise pine plantations.

Technology development history

Over the past 20 years, hyperspectral imagers already existed, but these advanced technologies were only available to government organisations and significant-sized institutions such as NASA – or in Australia, large mining companies.

This dynamic began to shift when the miniaturisation of these devices took place, including technologies such as GPS and inertial measuring units. Devices that originally costed millions of dollars to produce are now much more accessible and can be operated with minimal knowledge. This revolution in hyperspectral imager development has allowed for broader application of the technologies.

In a laboratory around 20-25 years ago, the very tiny chlorophyll fluorescence signal was first observed – consisting of about two per cent or so of the total incoming radiation. Since then, researchers have paid close attention to the monitoring and measuring of this signal because of its direct link with the photosynthesis of any vegetation worldwide at different scales.