Rapid changes in consumer preferences for high-quality animal-based protein have driven the poultry industry to identify non-invasive, in-line processing technologies for rapid detection of muscle meat quality defects. At production plants, technologies like radio-frequency waves (RF waves) can identify and separate myopathy-conditioned meat, reducing misclassification errors due to human fatigue and inexperience. Previous studies have shown that advanced diagnostic tools combined with complex data analytics, such as support vector machines (SVMs) and backpropagation neural networks (BPNNs), can classify chicken breast myopathies post-deboning. This study demonstrates RF wave use for myopathy detection at four processing stages. Using 107 (48-day old) broilers, RF wave data in amplitude and phase were collected from live birds, pre-chilled without giblets (WOGs), post-chilled WOGs, and freshly deboned fillets (3-3.5h post-slaughter) and examined by hand-palpation for woody breast categories (1-normal; 2-moderate; 3-severe). Data preprocessing involved false discovery rate and predictor analysis to identify specific signature frequencies and develop classification models using supervised machine learning (ML) algorithms. Variable clustering analysis identified seven to eight different frequencies at various processing stages. Preprocessed data with identified signature frequencies were used to develop classification models using BPNN and SVM. BPNN demonstrated superior classification accuracy compared to SVM, with accuracy ranges from 90.0% to 96.1% for live birds, 78.9% to 97.1% for pre-chilled WOGs, 82.1% to 95.9% for post-chilled WOGs, and 94.2% to 98.2% for deboned fillets. Integrating specific RF range devices or sensors with supervised ML algorithms like SVM and BPNN in poultry processing can effectively detect muscle myopathies at different processing steps during in-line processing.
Read full abstract