Multispectral cameras target industrial inspection

In recent decades, manufacturers have been faced with growing demand for increased production levels, reduced costs, and improved quality. To date, many of these applications have used monochrome vision systems to extract information such as object shape and placement. As these systems have proven effective, interest in the use of machine vision for more complex inspection tasks has grown. By imaging red, green, and blue spectral bands that match the sensitivity of the human eye, color vision sy

Apr 1st, 1999

Multispectral cameras target industrial inspection

In recent decades, manufacturers have been faced with growing demand for increased production levels, reduced costs, and improved quality. To date, many of these applications have used monochrome vision systems to extract information such as object shape and placement. As these systems have proven effective, interest in the use of machine vision for more complex inspection tasks has grown. By imaging red, green, and blue spectral bands that match the sensitivity of the human eye, color vision systems can analyze an image based on spectral effects.

While color cameras image light from 450 to 650 nm, the CCD sensors used in them can detect light from 400 to 1100 nm. Because of this, multispectral CCD cameras can add one or more wavebands beyond the 450-650-nm range, enabling new capabilities in automated food- and paper-inspection systems.

"In vegetation inspection," says David Duncan, president of Duncan Technologies Inc. (DTI; Auburn, CA), "the light reflected varies widely as a function of wavelength, especially in the near-infrared (NIR)." At the US Geological Survey Spectroscopy Laboratory (Denver, CO), the spectral reflectance of vegetation in the visible and short-wavelength NIR (SWNIR) regions has been measured. As plants typically exhibit a high reflectance in the SWNIR with a slight dip in the 940-1000-nm region, their relative reflectivities in various spectral regions can be used to interpret plant physical condition.

"Similar conditions can be observed in other organic materials," says Duncan. For example, the pigmentation in the skin of a red delicious apple produces the dark red appearance to the human eye. Yet the same pigmentation has virtually no effect in a SWNIR image. "Be cause the apple surface exhibits little absorption in the SWNIR, the image appears very light with virtually no artifacts relating to pigmentation. Similarly, the growth ring bands in lumber are prominent in the visible spectrum and barely visible in the SWNIR," he adds.

To date, multispectral imaging has been performed primarily with area-array video cameras. Use of video cameras for multispectral imaging has the drawback that the images are interlaced so any movement in the field of view between the odd and even field acquisition results in artifacts in the image. "Worse," says Duncan, "standard video cameras often have features not desirable in advanced imaging systems including gamma correction, AGC, and offsets."

Because multispectral imaging encompasses a range of applications, no single camera configuration can satisfy the various uses. Instead DTI has developed a camera architecture that can be customized. With support for analog output, the DTI cameras use a color-sorting prism to separate the image into three spectral bands. Two dichroic surfaces between the three prism elements are coated to isolate user-specified spectral bands.

Depending on the application, the system can be configured with three monochrome arrays or two monochrome and one color array to provide three- or five-band output. Each of the three spectral bands can be configured to image any band within the CCD`s 400-1100-nm spectral-response range. If the bands are configured to match the RGB primary color bands, the camera operates as a color camera.

"Initial camera configurations include a 640 ¥ 480 progressive-scan area array and 1024- or 2048-pixel linear arrays," says Duncan, "al though the architecture will accommodate high-resolution area arrays up to 1280 ¥ 1024 resolution."

More in Image sensors