A Non-destructive Detection Classification Method for Chilled Mutton's Freshness Based on Improved LightGBM Model
E Jixiang
JIANG Xinhua
XU Ziyang
BAI Jie
LI Jing
WU Guodong
ZHAI Chengjun
Abstract：In order to achieve rapid and non-destructive detection and classification of the freshness of mutton,in this work,a raw dataset of chilled mutton's freshness was first constructed through integrating multiple freshness indicators with hyperspectral data.A variety of preprocessing techniques were applied to establish a Support Vector Regression model,through which the optimal preprocessing method was determined.Subsequently,a feature selection process was initiated via Random Forest-based Recursive Feature Elimination and an autoencoder for band selection on the basis of the preprocessed datasets.A Light Gradient Boosting Machine(LightGBM)was subsequently built on the selected feature dataset to detect the freshness of chilled mutton.Given the limitations of traditional LightGBM in handling high-dimensional and sparse spectral data,particularly the low classification performance and lack of generalization capability,a multi-strategy optimization method based on the Hybrid Bat Algorithm and Nutcracker Optimization Algorithm was introduced to improve LightGBM.The improved LightGBM was compared with classification methods such as the original LightGBM,XGBoost,Support Vector Machine(SVM),and Random Forest(RF),followed by validation on datasets of different scales.The results demonstrated that the multi-strategy improved LightGBM achieved a freshness grade classification accuracy of 0.988 1,resulting in a 3.57%improvement compared with the unimproved version,with weighted precision,weighted recall,and weighted F1 scores reaching 0.988 6,0.988 2,and 0.988 3,respectively,all outperforming other classification methods.Furthermore,when detecting freshness grades across datasets of different scales,the model exhibited a consistent accuracy within the range of 0.984 0 to 0.989 0,indicating its good generalization capability and robustness.In summary,the effectiveness of the proposed improved LightGBM model has been validated in this work,providing a novel approach for rapid and non-destructive detection and classification of chilled mutton's freshness.
Keywords：mutton freshnesshyperspectralfeature selectionmulti-indicatorlightGBM
Publication Date：2026-01-20
Online Publishing Date：2026-03-31(First online date of this platform, not the publication date of the document)
Pages：11( 310-320 )