Fresh Meat Classification Using Laser-Induced Breakdown Spectroscopy Assisted by LightGBM and Optuna
Kaifeng Mo,
Yun Tang,
Yining Zhu,
Xiangyou Li,
Jingfeng Li,
Xuxiang Peng,
Ping Liao,
Penghui Zou
Affiliations
Kaifeng Mo
Hunan Province Key Laboratory of Intelligent Sensors and Advanced Sensor Materials, School of Physics and Electronics Science, Hunan University of Science and Technology, Xiangtan 411201, China
Yun Tang
Hunan Province Key Laboratory of Intelligent Sensors and Advanced Sensor Materials, School of Physics and Electronics Science, Hunan University of Science and Technology, Xiangtan 411201, China
Yining Zhu
Wuhan National Laboratory for Optoelectronics (WNLO), Huazhong University of Science and Technology, Wuhan 430074, China
Xiangyou Li
Wuhan National Laboratory for Optoelectronics (WNLO), Huazhong University of Science and Technology, Wuhan 430074, China
Jingfeng Li
Hunan Province Key Laboratory of Intelligent Sensors and Advanced Sensor Materials, School of Physics and Electronics Science, Hunan University of Science and Technology, Xiangtan 411201, China
Xuxiang Peng
Hunan Province Key Laboratory of Intelligent Sensors and Advanced Sensor Materials, School of Physics and Electronics Science, Hunan University of Science and Technology, Xiangtan 411201, China
Ping Liao
Hunan Province Key Laboratory of Intelligent Sensors and Advanced Sensor Materials, School of Physics and Electronics Science, Hunan University of Science and Technology, Xiangtan 411201, China
Penghui Zou
Hunan Province Key Laboratory of Intelligent Sensors and Advanced Sensor Materials, School of Physics and Electronics Science, Hunan University of Science and Technology, Xiangtan 411201, China
To enhance the accuracy of identifying fresh meat varieties using laser-induced breakdown spectroscopy (LIBS), we utilized the LightGBM model in combination with the Optuna algorithm. The procedure involved flattening fresh meat slices with glass slides and collecting spectral data of the plasma from the surfaces of the fresh meat tissues (pork, beef, and chicken) using LIBS technology. A total of 900 spectra were collected. Initially, we established LightGBM and SVM (support vector machine) models for the collected spectra. Subsequently, we applied information gain and peak extraction algorithms to select the features for each model. We then employed Optuna to optimize the hyperparameters of the LightGBM model, while a 10-fold cross-validation was conducted to determine the optimal parameters for SVM. Ultimately, the LightGBM model achieved higher accuracy, macro-F1, and Cohen’s kappa coefficient (kappa coefficient) values of 0.9370, 0.9364, and 0.9244, respectively, compared to the SVM model’s values of 0.8888, 0.8881, and 0.8666. This study provides a novel method for the rapid classification of fresh meat varieties using LIBS.
