Identification and classification of broiler carcasses exhibiting septicemia-toxemia using spectral imaging systems

Authors: BLACK, MICAH - Auburn University; GUZMAN, LUIS - Auburn University; SIDDIQUE, AFTAB - Fort Valley State University; SIERRA, KATHERINE - Auburn University; TASHIGUANO, VIANCA - Auburn University; GARNER, LAURA - Auburn University; MACKINNON, NICHOLAS - Safetyspect Inc; SOKOLOV, STANNISLOV - Safetyspect Inc; VASEFI, FARTASH - Safetyspect Inc; Baek, Insuck; Chao, Kuanglin; Kim, Moon; MOREY, AMIT - Auburn University

Submitted to: Food Control
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/30/2025
Publication Date: 7/3/2025
Citation: Black, M.T., Guzman, L., Siddique, A., Sierra, K., Tashiguano, V., Garner, L., MacKinnon, N., Sokolov, S., Vasefi, F., Baek, I., Chao, K., Kim, M.S., Morey, A. 2025. Identification and classification of broiler carcasses exhibiting septicemia-toxemia using spectral imaging systems. Food Control. 179. Article e111532. https://doi.org/10.1016/j.foodcont.2025.111532.
DOI: https://doi.org/10.1016/j.foodcont.2025.111532

Interpretive Summary: Currently, each chicken intended for sale to U.S. consumers is required by law to be inspected post-mortem by a USDA/FSIS (United States Department of Agriculture/Food Safety and Inspection Service) inspector for its wholesomeness. The FSIS inspectors visually examine the carcass exterior, the inner surfaces of the body cavity, and the organs of each carcass for indications of diseases or defects. For example, one such carcass defect that results in condemnation categorization in poultry inspection is called septicemia-toxemia. For effective inspection and occupational considerations, each inspector is limited to a maximum inspection speed of 35 birds per minute. This current inspection system limits the maximum possible production output of processing plants that are seeking to satisfy increasing consumer demand for U.S. poultry products. One possible solution to this problem is for poultry processing plants to install online instrumental inspection systems that can accurately screen wholesome carcasses for defects. In this study, we investigated the use of the handheld fluorescence imaging technology combined with machine learning methods to detect broiler septicemia-toxemic carcasses in laboratory settings. Results indicated that fluorescence imaging combined with machine learning models were able to classify all images at 100% accuracy. Although the system is currently handheld, further testing on poultry processing lines for real-world evaluation will give more accurate results on how effective the system could be when implemented on a commercial broiler processing line. The integration of imaging technology and Artificial Intelligence (AI) will benefit the U.S. poultry processing industry in the adoption of higher line speeds allowing for greater output and efficiency while maintaining food safety standards.

Technical Abstract: Septicemia-toxemia (sep-tox) is a bacterial infection in the bloodstream of live broilers resulting in red carcasses, dehydrated skin, or organ hemorrhaging. During processing, trained plant personnel visually inspect broiler carcasses separating those exhibiting sep-tox characteristics, which can be a fatiguing process. With the potential of increasing processing line speeds, it is appropriate to develop innovative real-time technologies to detect sep-tox. The investigation with RGB color imaging spectral camera system combined with machine learning for detection of sep-tox carcasses. Trained plant personnel identified a total of 380 sep-tox carcasses and 286 non-septox (market ready) carcasses and procured from a commercial poultry processor in twelve trials. Broiler carcasses were placed in an ambient-free light cabinet to prevent interference. The CSI-D+ imaging system collected two images per carcass at different exposure times to obtain a total of 1,332 images. Image analysis using data analytics was conducted with a supervised learning network, convolution neural network, to categorize images into normal and sep-tox. Two types of convolution neural network models classified with 100% testing accuracy average using the best produced model. RStudio used sep-tox images to receive a scatterplot showing diversity between sep-tox images. The scatterplot resulted in close knit clusters. Clusters were distinguishable resulting in 252 images in cluster one, 146 images in cluster two, and 214 images in cluster three. The spectral camera was effective detecting broiler septicemia-toxemic carcasses in laboratory settings and can be further investigated for application in the processing setting to aide with detection methods.