Incorporating Bioimpedance Technique with Ensemble Learning Algorithm for Mutton Tenderness Detection

To elucidate the role of bioimpedance technique in meat quality detection, we measured the impedance phase and modulus of chilled mutton with different storage time and temperature using 2-electrode and 4-electrode, respectively, and then generated the high-resolution impedance map. Nevertheless, tenderness, an important standard for the meat quality, is difficult to detect by relying only on electrode materials and impedance measurement approaches due to its nonlinearity and fuzziness. To overcome this challenge, we proposed a mutton tenderness detection method that incorporates the bioimpedance technique with an ensemble learning algorithm to improve the performance. This approach utilizes the advantages of multiple classical machine learning algorithms, such as SVM, ANN, and random forest, from a data-driven perspective. Importantly, we also introduced the lasso method to find significant impedance features that are more effective in improving the accuracy of the algorithm. The results showed that the stacking ensemble learning-based model exhibits the highest performance with an accuracy of 0.960, 0.986, and an F1-score of 0.969, 0.978 for 2- and 4-electrode, respectively, which are much higher than that of single machine learning algorithm. In conclusion, the proposed method demonstrated that ensemble learning algorithm can significantly improve the accuracy and efficiency of mutton tenderness detection. Furthermore, it also indicated that improving the model algorithm is also an important direction to promote the meat quality detection.