Towards multispectral imaging

Machine vision systems are now commonplace in many industrial inspection applications. However, this wasn't always the case. When the first machine vision systems were deployed in the 1980s, they were available from just a handful of vendors. These systems were expensive, often contained proprietary software and hardware, and were often rather unreliable.

Much has changed since then. Today, with a firm understanding of the application to be solved, systems integrators can choose from an array of cameras, interfaces, lighting and software from hundreds of vendors with which to develop machine vision systems. Needless to say, developing such systems from such OEM components results in systems that are more reliable and cost a fraction of what they did decades ago.

Among the different types of sensors and cameras available to meet challenging requirements are UV-based cameras for semiconductor inspection, visible light-based systems for industrial inspection, and IR detectors for medical and military applications. As the price of such detectors and cameras continues to fall, systems integrators will deploy them in more demanding multi-spectral applications.

Visible and IR systems can, for example, be used to detect whether a cap is properly fitted on a detergent bottle while at the same time examining whether the bottle is filled correctly. In security and surveillance systems, combining such multi-detector systems with sophisticated tracking software allows law enforcement officials to automatically track potential suspects.

In many of these applications, however, different types of cameras and detectors must be used to detect these various wavelengths, increasing their cost. Realizing this, a number of companies are now building detectors and cameras that can span wider wavelengths.

To do so, some vendors have incorporated two or three different types of detectors in their camera designs, allowing multi-spectral images to be captured with one single camera head. Also becoming available are CMOS-based imagers and cameras that span the visible to near infrared spectrum. While the spectral response of such detectors is not as sensitive in the infrared as their InGaAs counterparts, they are proving useful in applications where highly sensitive IR imaging may not be required.

Much of this development begs the question of whether true multi-spectral imagers could span larger ranges of the electromagnetic spectrum. Simple incarnations of such devices may first use filters similar to Bayer patterns to capture limited wavelengths at multiple frequencies. Other more complex implementations may use a combination of Group 5-7 elements. Just as machine vision systems were costly over 30 years ago, however, such detectors and the cameras based around them would initially be expensive and limited to military applications.