Growth position discrimination of flue-cured tobacco leaves based on hyperspectral imaging
[Background]Using hyperspectral imaging technology combined with machine learning methods,a recognition model for the growth potion(upper,middle,lower)of flue-cured tobacco was established.[Methods]The intensity distribution characteristics of tobacco leaves in water and protein sensitive bands was analyzed firstly.Then,a two-threshold region of Interest(ROI)selection method combining OTSU and Sauvola image segmentation algorithms was raised based on the analysis of abundance distribution.Moreover,the influences of different data preprocessing methods on data modeling were comparatively analyzed.Support vector machine(SVM)and extreme gradient boosting(XGBoost)algorithm were adopted to establish the discrimination model,and the models were optimized by parameter optimization subsequently.Using the Genetic Algorithm(GA)and the combination of Genetic Algorithm with Successive Projection Algorithm(GA-SPA)for the selection of characteristic wavelengths,a simplified model was established.[Results]Test data show:1)The established dual threshold method can efficiently select the right leaf area of flue-cured tobacco leaves;2)the efficacy of discrimination model was obviously affected by data preprocessing methods.Based on the spectral data preprocessed using the first-order derivative and Savitzky-Golay smoothing(1Der+SG),combined with the characteristic wavelengths selected by GA,the XGBoost recognition model for the growth position had the best classification performance,with an accuracy rate of up to 97.78％.[Conclusion]The model established in this study,based on hyperspectral imaging technology combined with machine learning methods,can efficiently and accurately identify the growth position of flue-cured tobacco.
万方数据
维普
