Submitted to: ASAE Annual International Meeting
Publication Type: Proceedings
Publication Acceptance Date: 6/27/2003
Publication Date: 7/27/2003
Citation: Park, B., Lawrence, K.C., Windham, W.R., Smith, D.P., Feldner, P.W. 2003. Machine vision for detecting internal fecal comtaminants of broiler carcasses. American Society of Aricultural Engineers Annual International Meeting. St. Joseph, MI. Technical Paper No.033051. Interpretive Summary: An imaging system can be effectively used for detecting internal fecal contaminants of a broiler carcass and demonstrates potential application for poultry safety inspection. Previously, hyperspectral imaging techniques have been applied to detect fecal contaminants on the surface of poultry carcasses. In order to successfully implement an imaging system for poultry fecal inspection, internal fecal contaminants must also be detected. However, vision techniques for detecting fecal contaminants in the visceral cavity of broiler carcasses are challenging. We employed the state-of-the-art hyperspectral imaging system to detect fecal contaminants in the visceral cavity of carcass halves. The preliminary study showed that the image processing methods, specifically image band ratio of selected wavelength images and threshold, which separates fecal spots from the surface, performed well to identify fecal contaminants in the visceral cavity of broiler carcasses. The selected image processing methods from the hyperspectral images can be further applied for real-time detection of fecal contaminants in the visceral cavity of broiler carcasses.
Technical Abstract: Detecting fecal contaminant in the visceral cavity of the broiler is difficult but extremely important for poultry safety inspection. A hyperspectral imaging system could be used effectively for detecting internal cecal contaminants. A hyperspectral imaging system was used to image broiler carcass halves with cecal contaminants. Two 565 and 517-nm wavelength images were selected from 512 calibrated hypercube image data. Image processing algorithms, band ratio, threshold, and median filtering, were useful to identify fecal contaminants from the internal cavity. The accuracy of detection algorithms to identify cecal contaminants was varied with fecal threshold values and median filter as well. The imaging system easily identified cecal contaminants with 92.5% detection accuracy but also incorrectly identified 123 carcass features that were not considered as contaminants (false positives) and missed 15 actual contaminants (7.6% Type I error) when fecal threshold value of 1.05 was employed. The higher accuracy (96.9%) and lower missed contaminants (3.0% Type I error) could be obtained when different fecal threshold values were used. However, in this case, false positives drastically increased.