|Chao, Kuanglin - Kevin Chao|
Submitted to: Transactions of the ASAE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/20/2000
Publication Date: 10/20/2000
Citation: N/A Interpretive Summary: The current organoleptic method is labor-intensive and is a limiting factor for higher poultry productive capacity. Due to massive production of poultry and poultry products and the inherent variability and complexity in individual birds, there is a great need to improve the existing inspection methods. In this study, two dual-camera systems were developed for on-line inspection of poultry carcasses, one imaging the front of the carcass, the other the back. Each system was designed, fabricated, and evaluated in an in-house pilot scale poultry processing line. Each system has illumination, two cameras with filters of different wavelengths, and an image acquisition, control, and analysis modules. Systems level analysis and design integrated individual vision hardware components with image processing algorithms. The front imaging system gave accuracies of 91%, 98%, and 95% for normal, abnormal, and combined carcasses, respectively. The back imaging system gave 84%, 100%, and 92%. This information is useful to the Food Safety and Inspection Service (FSIS) and the poultry processing industry.
Technical Abstract: Two dual-camera systems were developed for on-line inspection of poultry carcasses: one to image the front of the bird and the other to image the back. Each system consists of two identical black and white cameras equipped with interference filters of 540 nm and 700 nm. Both cameras capture spectral images simultaneously. Object-oriented analysis was performed to identify the attributes of individual software components and the relationships among these software components. These individual software components were then organized by the object patterns to form a software architectural framework for on-line image capture, off-line development of classification models, and on-line classification of carcasses into wholesome and unwholesome categories. Model development and testing was performed on 331 chickens independently classified by a veterinarian. For off-line model development, the accuracies for differentiating between wholesome and unwholesome carcasses were 96.2% and 88.5% at 540 nm and 700 nm, respectively, for the front images and 95.7% and 85.1% at 540 nm and 700 nm, respectively, for the back images. On-line classification for 128 new samples combined the filter information within each system, using the optimal neural network models. The front imaging system gave accuracies of 91%, 98% and 95% for normal, abnormal and combined carcasses, respectively. The back imaging system gave 84%, 100% and 92%.