Application of XGBoost-RFE-SHAP in the selection of spectral feature wavelength: A case study of in situ detection for typical foodborne pathogens contamination on mutton

文献类型： 外文期刊

第一作者： Bai, Zongxiu

作者： Bai, Zongxiu;Xing, Fukang;Hu, Yuxia;Zhu, Rongguang;Bai, Zongxiu;Zhu, Rongguang;Du, Dongdong;Kang, Lichao

作者机构：

关键词： Mutton; Foodborne pathogens; HSI; Feature wavelength; XGBoost-RFE-SHAP

期刊名称：FOOD CONTROL （ 影响因子：6.3； 五年影响因子：6.1 ）

ISSN： 0956-7135

年卷期： 2025 年 177 卷

页码：

收录情况： SCI

摘要： To solve the problems of many feature wavelengths and poor interpretation in detecting foodborne pathogens contaminated on mutton by HSI, the eXtreme Gradient Boosting (XGBoost)-Recursive Feature Elimination (RFE)-SHapley Additive exPlanations (SHAP) was proposed to reduce and explain wavelengths of Visible Near-Infrared (VNIR)-Hyperspectral Imaging (HSI) and Short-Wave Infrared (SWIR)-HSI. The influence of different preprocessing methods on the model was discussed. Based on the preprocessed spectra, the feasibility of establishing simplified models for detection foodborne pathogens contaminated on mutton by using the feature wavelength selected by XGBoost-RFE-SHAP method was discussed. The results showed that second derivative preprocessing could improve the accuracy of Long Short-Term Memory (LSTM), but it had no effect on One-Dimensional Convolutional Neural Network (1D-CNN) model. The number of feature wavelengths selected by XGBoost-RFE-SHAP in VNIR and SWIR band was to 28 and 19, accounting for 5.73 % and 8.52 % of the whole band, respectively. The simplified detection models were established by feature wavelengths combined with LSTM. The results showed that test set and external validation set were 88.39 % and 85.71 % in VNIR band, respectively, and 91.07 % and 91.07 % in SWIR band, respectively. The results indicated that XGBoost-RFE-SHAP was an effective method to select the feature wavelength from HSI, and detection models with good performance under the condition of a small number of features were established based on LSTM model. The results provide theoretical basis and technical support for the development of multi-spectrometer for detection of foodborne pathogen contamination on mutton.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序