Multispectral and Hyperspectral Imaging Delivers Precision Inspection
| By: Winn Hardin, Contributing Editor
From inspecting PCBs on the production line all the way up to crop observation via satellite, multispectral and hyperspectral imaging allows users to detect very small differences that traditional RGB or monochrome cameras alone cannot.
The technique is attracting more users, as the global hyperspectral imaging market is expected to reach $74.29 million by 2019 with a growth rate of 12% from 2014 to 2019. And, thanks to advances in sensors, filters, and processors, multi- and hyperspectral imaging is becoming more affordable, compact and accessible for a variety of applications.
Inspection From the Skies Down
Remote sensing represents one of the biggest applications for multi- and hyperspectral imaging. “Some of these filters are specifically designed to look at forestry, for example,” says Jean Pierre Luevano, international sales manager at Teledyne DALSA (Waterloo, Ontario, Canada). “If there is a satellite flying over Asia, you will have a very specific filter that filters out clouds clear from the sun to look at a particular wavelength where you can very clearly see rice paddies, for example.”
Teledyne DALSA is developing a new multispectral image sensor for Denel Spaceteq, a South Africa–based government agency that provides high-performance satellite systems. The sensor will enable advanced earth observation applications, among them vegetation. “You can imagine that countries in Africa, where you have growing sand dunes and shrinking vegetation that governments would want to measure the change between the vegetation and the arid area, and then form a plan to respond to either stop one or improve the other,” Luevano says.
Government agencies and agricultural companies alike are deploying multispectral and hyperspectral imaging to measure crop health and growth. That could entail anything from identifying water-stressed plants to detecting diseased crops in the field. “Strawberry growers have an interest in using multispectral analysis to discern sugar content in the berries because after they’re picked the growers want them to be sorted for high quality for direct consumption as fruit, versus lower quality that goes into juices or additives,” says Dr. Rex Lee, president of Pyramid Imaging (Tampa, Florida).
More frequently, multispectral imaging for precision agriculture applications is finding a home on unmanned aerial vehicles (UAVs) as cameras become smaller. Ximea (Marianka, Slovakia) is set to release the MQ022HG-IM-LS150-VISNIR, the most recent line-scan hyperspectral camera in its xiSpec series. Covering the visible and near infrared (NIR) in one package, the camera is based on a sensor from Belgium-based imec, which has developed a process that applies spectral filters at the pixel level on image sensors. The sensor features 150+ bands in the 470-900 nm range. The result is a camera that weighs only 32 g with low power consumption.
“For a line-scan system, you have to move the camera or the object or both, and at least 150 pictures must be stitched together for the complete hyperspectral information,” explains Ximea COO Jürgen Hillmann.
In addition to the agricultural industry, Hillmann says the MQ022HG-IM-LS150 is suitable for more local or mobile handheld systems, including medical applications such as skin detection to test blood oxygenation levels, detecting piracy on banknotes with non-invisible markers, and food sorting.
Multispectral imaging is just as useful on the plant floor. Teledyne DALSA initially developed its Piranha4 2k quadlinear CMOS line-scan multispectral camera for machine vision applications. The camera features RGB outputs plus a monochrome or near-infrared (NIR) channel that allows not only electronics manufacturers but also OEMs in print inspection and food and material sorting to detect defects they couldn’t see before.
Meanwhile, Adimec (Eindhoven, The Netherlands) has partnered with imec to assess the capability of instant hyperspectral imaging at video rates and faster with its customers in machine vision, healthcare, and global security. Adimec’s Quartz camera series serves as the evaluation platform to facilitate hyperspectral cubes with a 2-megapixel field-of-view at 340 frames per second.
On the Path to Wider Adoption
For all the advantages it provides, multispectral and hyperspectral imaging still contends with limitations. Whereas multispectral uses a handful of spectral bands, hyperspectral can use hundreds. “We can put filters in front of a camera and get wavelength dispersion across the sensor, and we can make that camera a hyperspectral data collector,” Dr. Lee says. “But when looking out into a complex scene, you’re going to see a multitude of spectrum that’s going to be a conglomeration of everything in the scene. And that gets quite complicated.”
As a result, hyperspectral imaging is data intensive, requiring a lot of computing power and bandwidth to transmit the vast amounts of information collected by the camera. Ximea hopes to address some of these problems by extending its offerings from cameras to computational modules capable of processing hyperspectral data in real-time.
“The cameras will not just be standalone but rather embedded where the camera is tightly connected to a computational engine and instead of broad images delivers actionable data,” says Max Larin, CEO of Ximea.
To illustrate his point, Larin points to the NVIDIA® Jetson™ TX1, a credit card–sized embedded supercomputer designed for drones. The module, which offers 1 Tflop of performance, uses machine learning in tandem with vision to recognize objects or interpret information.
“We are creating carrier boards that include all the interconnections, which covers not only power supply and data but synchronization and respective software,” Larin says. “And our partners are integrating the hardware into the enclosure, since the components are lightweight and only use up to 6 or 7 W of power when running full speed.”
When it comes to multi- and hyperspectral imaging for space applications, cost remains one of the biggest barriers to entry for high-resolution imagers. “Multispectral imaging is not used more widely due to the fact that these detectors do tend to be expensive,” says Teledyne DALSA’s Luevano. “Typically, smaller countries cannot afford to pay for a full-fledged payload, or full satellite. That can run into hundreds of millions of dollars.” In these instances, smaller governments or private entities end up purchasing the data rather than developing their own remote-sensing system.
Although multispectral and hyperspectral imaging still encounters challenges in broader adoption of the technology, shrinking camera sizes and better processing capabilities are opening the door to more applications that require precision inspection.