A camera which captures images across a wide range of wavelengths, not just the familiar RGB colours has existed for a while now. Lots of niche applications have been found – from the military looking for camouflage amongst vegetation to conservationists finding obscured details in Renaissance paintings.
Semiconductor and imaging products have moved forward at an astounding pace – but why is such compelling technology not on the wider market – or on a phone?
The progress in 2D imaging sensors has taken massive investment, fuelled by the uptake rate of smartphones. This has funded dozens of engineering cycles, each making competitive improvements and reducing cost. However, except for niche applications, hyperspectral imaging hasn’t found a ‘killer app’ that will bring costs and complexity down to a mass market level.
A good test of whether technology is mass market ready is whether it’s made its way down to low margin areas such as food and farming – and indeed the principles are there.
It’s clear that plants can change colour when unhappy – hyperspectral imaging is able to probe deep into this and see signs of water stress or fungal infection before they become visible to the naked eye. This is due to changes in the cell walls which are only visible in the infra-red. The changes happen at a large enough scale to be captured by a camera on a drone.
Although this is a breakthrough, it’s not clear how this will fit into a busy farm. There is quite a chain here – find a pilot, fly the fields, reprocess the images, extract the actionable information and then apply the right treatment. This is a typical ‘Swiss Army Knife’ problem, the tool is too general purpose for the job and becomes too much trouble to use. The solution, of course, is to specialise on a particular task with a clear business case.
This specialisation for a particular task allows the engineers to focus on a more efficient solution. This could be reducing the cost (sometimes £10,000 for a general purpose machine) dramatically and speeding up the imaging process. We have done this for two applications areas: yield forecasting for crops and developing chemical sensitivity for medical and industrial applications.
Yield forecasting is accomplished by exploiting the contrast between different types of leaves, or fruits and leaves. This contrast can be brought out by examining just a few wavelengths – this reduces cost but importantly increases the scan speed to a feasible level.
Chemical sensitivity is achieved in the opposite way – by increasing the detail in which wavelength data is captured. The challenge is in imaging at a good scan speed – we have achieved this by simplifying the optics in a radical way. This can unlock new applications which need imaging and the kind of discrimination only available on a benchtop spectrometer.
The focus on a customer problem has inspired engineers to think again about what hyperspectral imagers need to look like – finally they may be worth the hype.