Submitted to: Food Safety International Conference
Publication Type: Abstract Only
Publication Acceptance Date: August 1, 2006
Publication Date: August 22, 2006
Citation: Lefcourt, A.M. 2006. Use of imaging and spectroscopy for non-destructive food safety inspection in real time [abstract]. Food Safety International Conference, August 22 - 25, 2006, Guadalajara, Mexico.
The Instrumentation and Sensing Laboratory (ISL) has a long history of using imaging and spectroscopy to address agricultural problems. Examples include identification of normal and scab damaged wheat kernels, and wholesome and unwholesome chicken carcasses. Scientists at ISL develop the theoretical and practical methods needed to solve a problem, and carry the work through to construction of a commercially-viable detection system. A recent interest concerns the problem of detecting fruits and vegetables contaminated with feces. Feces is the root source of most harmful microorganisms in foods. A hyperspectral imaging system was used to examine whether feces applied to normal apples could be detected using reflectance or fluorescence imaging techniques. It was determined that detection was feasible using either technique, but that sensitivity was better using fluorescence. The fluorescence response of feces was used to develop a practical detection system, but only after a number of confounding issues were resolved. The primary issue is that chlorophyll and chlorophyll-related compounds are responsible for the fluorescence response of feces, and apples contain large quantities of chlorophyll. Furthermore, quantities vary significantly across the surface of individual apples and between apples. Two techniques were developed to address this variability. Ratios of images acquired at two different wavelengths when used with an intensity-compensated threshold can normalize the effects of this variability. Alternatively, a technique was developed to normalize the power in individual images of apples. Another issue is the interrelated factor of cost, computational speed, and the need to process images at line processing speeds of 10 apples/s. This constraint suggests that it would be preferable to use multi-spectral rather than hyperspectral imaging in a practical detection system, and that use of one or two wavelengths would be best. A number of studies were conducted to determine optimal wavelengths for detection; an ancillary prerequisite for this work was the development of novel optimization algorithms. Peaks and valleys in weighting functions for principal component images that showed good visual contrast between contamination sites and normal apple surfaces yielded candidate wavelengths for further testing. Once a limited number of detection algorithms were identified, an exhaustive search routine allowed detection algorithms to be optimized in terms of numbers of false positives. Results showed that use of two wavelengths was both required and adequate for detection of feces on apples. As the fluorescence response is weak, a pulsed-laser was used as the excitation source to allow responses to be detected under ambient lighting conditions, i.e., not in the dark. Use of a pulsed laser with a fast-gated intensified camera permitted use of ns-scale time-resolved imaging, and time resolved imaging was found to allow the most sensitive and selective detection of feces on produce. A practical detection system was developed using a commercial apple sorter. The system uses a high-intensity UV source, a prism-based spectral adapter, and a camera with an electron-multiplying detector (CCD). The camera allows spectral information to be binned as desired, and this binning capability is used to capture line-scan images at two selected wavelengths and wavebands as the apples pass under the camera. Detection of apples artificially contaminated with feces is 100% with no false positives.
Development of a commercial detection system requires a means to appropriately orient apples so that the entire surface of an apple can be imaged. Preliminary tests demonstrated that apples could be oriented along the stem/calyx axis using inertial properties. When apples were rolled down two parallel tracks, apples moved to an orientation where the stem/calyx axis was perpendicular to the direction of travel and parallel to the plane of the tracks after sufficient angular velocity was achieved. The potential for using this phenomena to orient apples for imaging is currently under investigation.