Hyperspectral cameras split the light reflected by an object into many narrow spectral bands, which they capture and process separately. That way, they record a spectral signature for each pixel in a scene. This signature .
Hyperspectral cameras split the light reflected by an object into many narrow spectral bands, which they capture and process separately. That way, they record a spectral signature for each pixel in a scene. This signature is much richer than the red-green-blue image that our eyes capture. It may even uniquely identify the material in the picture since each molecule interacts with light in a specific way, resulting in a spectral fingerprint. These unique fingerprints are of excellent value for identifying and classifying all sorts of materials and objects, and they are vital for further automation in industrial processes. It allows us to, e.g., reveal the corrosion on the structural elements of a bridge.
The most common implementation of these cameras is called a « push-broom, » which scans a scene line by line and takes several seconds or even minutes. Furthermore, these cameras are assembled by hand from many discrete components, including expensive and heavy precision optics in the glass. It requires careful alignment and calibration. These implementation factors mean only experts in hyperspectral imaging can calibrate, correct, and interpret the hyperspectral data into actual industry solutions.
Push-brooms do a great job in a conveyor-belt situation but have difficulties with a camera or scene in free movement. Think of an inspection camera in a probe robot or a free-flying drone. For these, you need to scan whole frames, preferably at video speed, to observe real-time changes. For example, when a robot takes random items out of a bin, it must take pictures permanently to make new decisions after each action. When the image gets refreshed at a rate of, e.g., 30 frames per second, this results in a hyperspectral video. Today, hyperspectral video is ready to use in a wide variety of industrial contexts: quality inspection, sorting, and material detection. So how did we get there? The short answer: is chip technology.
Using CMOS-based infrastructure and process technology in a cleanroom, imec developed a chip with integrated hyperspectral functionality by building interference-based optical filters at the wafer level, depositing and patterning them directly on top of image sensor pixels.
