Skip to main content
ARS Home » Research » Publications at this Location » Publication #190609


item Lefcourt, Alan
item Kim, Moon
item Chen, Yud

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/28/2006
Publication Date: 2/1/2007
Citation: Lefcourt, A.M., Kim, M.S., Chen, Y.R., Kang, S. 2006. Systematic approach for using hyperspectral imaging data to develop multispectral imagining systems: detection of feces on apples. Computers and Electronics in Agriculture. 54:22-35.

Interpretive Summary: Hyperspectral data sets contain a number of images of the same object at different wavelengths. Hyperspectral data is often used to identify a limited number of wavelengths that contain critical information about an object, such as whether the object is contaminated with a foreign substance. This reduction in the quantity of data that needs to be examined makes construction of commercial detection hardware feasible. A particular problem is fecal contamination of apples, a well recognized health safety issue. This paper examines methods of selecting one or two wavelengths from hyperspectral data sets with 60-120 wavelengths for the purpose of detecting feces on apples with minimal false positives. In the past, scientists have examined the characteristics of raw images to select the appropriate wavelengths to use for detection. In the current study, a range of potential detection methods were applied to the images, and the best waveforms were selected in terms of the actually ability to detect feces on apples. Results demonstrated that feces could be detected with minimal false positives using either reflectance of fluorescence images. Sensitivity was slightly better using the fluorescence images. The method developed for identifying appropriate wavelengths can be applied in many areas besides detecting feces on apples, and will be of interest to scientists and engineers faced with the task of reducing hyperspectral data to a more usable format.

Technical Abstract: The large size of data sets generated using hyperspectral imaging techniques significantly increases both the ability and difficulty of designing detection and classification systems. Of particular interest is the confluence with increasing use of multispectral imaging in machine vision, particularly in the area of food safety inspection. The purpose of this study was to develop a robust method for selecting one or two wavelengths for a multispectral detection system using hyperspectral data. The actual performance of detection algorithms in terms of true positives and false positives was used as optimization criteria. Detection of fecal contamination on apples is an important health safety issue. Prior observations suggest reflectance or fluorescence imaging in the visible to near-infrared can be used to detect such contamination. For this study, 1:2, 1:20, and 1:200 dilutions of dairy feces were applied to 100 Golden and 100 Red Delicious apples, and the treated apples were imaged using a hyperspectral system. Half of the apples were used to develop detection algorithms, and the remainder were used for validation. Test images were constructed from single band images and ratios or differences of images. A uniform power transformation was then used to reduce inter-apple intensity variability. Detection was accomplished using a binary threshold applied to transformed images or to transformed images subject to a 5 by 5 Prewitt edge detection filter. Thresholds for detection were optimized allowing for three false positives. A software routine was written to allow detected sites to be classified as one of the treatment sites or as a false positive. A second routine was used to iteratively construct images and to run tests. For reflectance imaging, difference images produced the best detection. For Golden and Red Delicious apples detection rates for 1:20 dilution spots were 100% and 62.5% using R816 - R697 and R784 - R738, respectively. For fluorescence imaging, ratio images produced the best detection. For Golden and Red Delicious apples detection rates for 1:200 dilution spots were 97.9% and 58.3% using F665/F602 and F647/F482, respectively. In all cases, more concentrated dilutions were detected at 100%. Detection rates for Red Delicious apples were improved by the use of the Prewitt filter. In the current study, detection rates using wavelengths identified in previous studies using statistical methods such as principal component analysis were lower in comparison, mainly due to problems with false positives. The procedures used for developing detection algorithms are not specific to detecting feces on apples, and it is theoretically easy to extend the results to detection schemes involving many wavelengths. The problem is the classical dilemma of rapidly increasing computational time. Still, given the costs of thoroughly testing a candidate detection algorithm, the time maybe warranted. Furthermore, as machine vision systems are often limited to one or two wavelengths due to practical considerations including cost, exhaustive search algorithms based-on optimizing the output of candidate detection algorithms should be cost effective.