Skip to main content
ARS Home » Southeast Area » Athens, Georgia » U.S. National Poultry Research Center » Quality and Safety Assessment Research Unit » Research » Publications at this Location » Publication #405453

Research Project: Smart Optical Sensing of Food Hazards and Elimination of Non-Nitrofurazone Semicarbazide in Poultry

Location: Quality and Safety Assessment Research Unit

Title: Semisupervised deep learning for the detection of foreign materials on poultry meat with near-infrared hyperspectral imaging

Author
item CAMPOS, RODRIGO LOUZADA - UNIVERSITY OF GEORGIA
item Yoon, Seung-Chul
item CHUNG, SOO - SEOUL NATIONAL UNIVERSITY
item BHANDARKAR, SUCHENDRA - UNIVERSITY OF GEORGIA

Submitted to: Sensors
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/4/2023
Publication Date: 8/8/2023
Citation: Campos, R., Yoon, S.C., Chung, S., Bhandarkar, S.M. 2023. Semisupervised deep learning for the detection of foreign materials on poultry meat with near-infrared hyperspectral imaging. Sensors. 23(16): 7014. https://doi.org/10.3390/s23167014.
DOI: https://doi.org/10.3390/s23167014

Interpretive Summary: This paper describes the development of a novel semi-supervised hyperspectral imaging technique for detecting foreign materials (FMs) on raw poultry meat during processing. FMs in poultry meat pose food safety concerns and financial losses for processors due to recalls. Conventional sensing technologies such as X-ray and metal detection have limitations in detecting low-density FMs, while color imaging technologies struggle with color ambiguity between meat and FMs with similar colors. Hyperspectral imaging combined with deep learning has shown promise in characterizing food safety and quality attributes, but it requires a large amount of accurately annotated data for supervised training, which is impractical for collecting diverse FM samples. To address this data collection problem, the paper proposes a semi-supervised hyperspectral deep learning model based on a generative adversarial network without requiring any FMs during model training. The model is trained using approximately 879,000 spectral responses from hyperspectral images of clean chicken breast fillets in the near-infrared wavelength range. For testing, 30 different types of FMs in two nominal sizes (2 × 2 mm2 and 5 × 5 mm2) were used. The FM detection technique achieved impressive results at both the spectral pixel level and the object level. At the spectral pixel level, the model achieved high precision (100%), recall (93%), F1 score (96.8%), and balanced accuracy (96.9%). When combining spectral data with spatial information, the FM detection accuracy at the object level reached 96.5%. With its ability to generate synthetic data and overcome the limitations of traditional sensing technologies, the proposed semi-supervised hyperspectral image-based deep learning paradigm holds potential for generalization and application in solving FM detection problems in various agriculture and food-related applications.

Technical Abstract: A novel semisupervised hyperspectral imaging technique was developed to detect foreign materials (FMs) on raw poultry meat. Combining hyperspectral imaging and deep learning has shown promise in identifying food safety and quality attributes. However, the challenge lies in acquiring a large amount of accurately annotated/labeled data for model training. This paper proposes a novel semisupervised hyperspectral deep learning model based on a generative adversarial network, utilizing an improved 1D U-Net as its discriminator, to detect FMs on raw chicken breast fillets. The model was trained by using approximately 879,000 spectral responses from hyperspectral images of clean chicken breast fillets in the near-infrared wavelength range of 1000–1700 nm. Testing involved 30 different types of FMs commonly found in processing plants, prepared in two nominal sizes: 2 × 2 mm2 and 5 × 5 mm2. The FM-detection technique achieved impressive results at both the spectral pixel level and the foreign material object level. At the spectral pixel level, the model achieved a precision of 100%, a recall of over 93%, an F1 score of 96.8%, and a balanced accuracy of 96.9%. When combining the rich 1D spectral data with 2D spatial information, the FM-detection accuracy at the object level reached 96.5%. In summary, the impressive results obtained through this study demonstrate its effectiveness at accurately identifying and localizing FMs. Furthermore, the technique’s potential for generalization and application to other agriculture and food-related domains highlights its broader significance.