REAL-TIME MULTISPECTRAL IMAGE PROCESSING FOR POULTRY INSPECTION

Authors: Park, Bosoon; Chen, Yud

Submitted to: Journal of Food Process Engineering
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/30/2000
Publication Date: N/A
Citation: N/A

Interpretive Summary: This research studied a real-time machine vision system for on-line poultry inspection using neural network image processing and analysis techniques. During the last three decades poultry production and the processing speed at slaughter plants has greatly increased. Therefore, fast, accurate, and reliable instrumental inspection tool is needed for improving the federal poultry inspection program. The Instrumentation and Sensing Laboratory at Beltsville, Maryland has devised a multispectral imaging system for instrumental poultry inspection. This imaging system scans whole body of each bird. The machine vision system was used to acquire spectral images from the chickens on a moving shackle with 60 birds/minute. Neural-network-based spectral image analysis could classify wholesome and unwholesome poultry carcasses with high accuracy. The multispectral imaging system can be used for eliminating condemned carcasses in an on-line system. This information is useful to the Food Safety and Inspection Service (FSIS) and to researchers who are developing machine vision systems for real-time grading or inspection of agricultural products.

Technical Abstract: A real-time multispectral image processing algorithm was developed for on-line poultry carcass inspection. Neural network models with different learning rules (delta and hyperbolic tangent) and transfer functions (sigmoid and norm-cum-sigmoid) were examined using features extracted from spectral images at 540 nm and 700 nm. The classification accuracy using dual wavelength spectral images was much higher than single wavelength spectral images in identifying unwholesome poultry carcasses. The spectral image features at 700 nm were useful to identify wholesome carcasses, while the combination of spectral image features at 540 nm, 700 nm, and their ratio improved the classification accuracy of unwholesome carcasses. The optimum neural classifier utilizing delta learning rule and hyperbolic tangent transfer function. Classification accuracy was 91.1% for wholesome and 83.3% for unwholesome carcasses when the spectral images of all 540 nm, 700 nm, and their ratio were used as inputs to the neural network model.