Submitted to: Applied Engineering in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: May 18, 1999
Publication Date: N/A
Interpretive Summary: Color is an important attribute for poultry inspection. In the poultry processing plant, USDA certified inspectors perform an inspection process, where most conditions of contaminated poultry are detected based upon variations in color. Usually, the inspection tasks are conducted at high rates (35 birds per minute or higher) making it a very difficult task for the inspectors. With the availability of improved hardware for acquiring color images, the capability now exists for development of color vision systems for on-line poultry inspection. This research was conducted to develop a neuro-fuzzy based image classification system to identify individual features of poultry viscera. Poultry viscera including livers and hearts were imaged in the poultry process plant. These images were segmented and statistical analysis was performed for feature selection. The color-image classification systems that utilized hybrid paradigms of fuzzy inference system and neural networks were designed for poultry viscera classifications. Results showed that color features could be effectively used for differentiating normal poultry viscera from airsacculitis, cadaver, and septicemia. For four-class classification of chicken livers and hearts, an accuracy of 86.3% was achieved for the model development and 82.5% accuracy for the model validation. The classifier design examples documented in this study illustrate the design procedures, which could be applied to the design of on-line systems for poultry viscera inspection. This information is also very useful to FSIS scientists and administrators who are interested in machine vision systems for poultry inspection.
A neuro-fuzzy based image classification system that utilizes color-imaging features of poultry viscera in the spectral and spatial domains was developed. Poultry viscera (320) of liver and heart were separated into four classes: normal, airsacculitis, cadaver, and septicemia. Color images for the classified poultry viscera were collected in the poultry process plant. These images in RGB color space were segmented and statistical analysis was performed for feature selection. The neuro-fuzzy system utilizes hybrid paradigms of fuzzy inference system and neural networks to enhance the robustness of the classification processes. The results showed that the accuracy for separation of normal from abnormal livers were 87.5 to 92.5% when two classes of validation data were used. For two-class classification of chicken hearts, the accuracies were 92.5 to 97.5%. When neuro-fuzzy models were employed to separate chicken livers into three classes (normal, airsacculitis, and cadaver), the accuracy was 88.3% for the training data and 83.3% for the validation data. Combining features of chicken liver and heart, a generalized neuro-fuzzy model was designed to classify poultry viscera into four classes (normal, airsacculitis, cadaver, and septicemia). The classification accuracy of 86.3% was achieved for the training data and 82.5% accuracy for the validation.