Hyperspectral imaging has identified soybean plants affected by sudden-death syndrome up to several days, and in some cases about a week, before the damage becomes visible to the human eye. In a greenhouse study at North Dakota State University, the diseased plants could be separated from healthy ones 12 days after planting, even though they still looked similar externally.
The result points to a possible agricultural sensing tool with clear practical value: early disease detection could give farmers time to intervene before damage spreads. The main obstacle is cost. Hyperspectral cameras can run into tens or hundreds of thousands of dollars, require specialized expertise, and produce very large data sets that are difficult to process in the field.
A key idea is to use hyperspectral measurements to identify only the wavelengths that actually distinguish diseased from healthy crops. Conventional cameras work with broad red, green, and blue bands, while hyperspectral systems divide light into many narrow bands and can reveal physiological or chemical changes long before they affect appearance. A short-wave infrared camera used in the research covers roughly 1000 to 2500 nm and more than 200 spectral bands.
Once the relevant wavelengths are known, a much simpler multispectral camera could capture just those bands. Multispectral systems with four to 10 bands already cost far less than research-grade hyperspectral instruments, and reducing the number of bands also reduces the computing load. In the intended farm use case, only three or four bands might be enough for a small processor to generate a result quickly enough for field work.
There are still major technical hurdles. Similar spectral changes can be caused by different stresses, especially before visual symptoms appear, and sunlight, soil conditions, and other environmental factors also affect measurements. The soybean work has not yet been validated under field conditions, and a sufficiently reliable wavelength set for field diagnosis has not yet been established.
At imec, researchers are pursuing sensor technologies that could lower the cost of application-specific spectral sensing. The work starts with high-resolution cameras that capture around 100 bands, then narrows the selection to the most relevant ones. For many applications, 10 or fewer bands are sufficient, and standardized sensors with nine, 16, or 25 bands could serve multiple uses while spreading manufacturing costs.
The fabrication approach matters as much as the optics. Instead of relying on separate components to split light into wavelengths, imec integrates spectral filters directly with image detectors during semiconductor fabrication. Because this is done at wafer level, many detectors can be manufactured at the same time. The result is a smaller, lighter sensor with fewer alignment requirements and the option for compact snapshot capture, which is useful when the sensor is mounted on moving tractors or drones.
Volume remains the final economic barrier, especially for short-wave infrared devices, which are still more expensive than conventional CMOS1. complementary metal-oxide semiconductorCircuit technology combining n- and p-channel MOS transistors for low static power consumption.2. complementary metal-oxide-semiconductorCMOS is a fundamental semiconductor technology for digital and analog circuits. sensors. Even so, the combination of wavelength selection, wafer-level integration, and higher production volumes could move spectral sensing from laboratory equipment toward farm machinery.