基于高光谱成像技术与GBDT的大豆品种无损鉴别

Non-destructive Identification of Soybean Varieties Based on Near Infrared Hyperspectral Technique and GBDT

大豆的品种直接关系到大豆制品的质量和出油率,目前主要采用对大豆中蛋白质及脂肪等含量的检测来实现对大豆品种的鉴别。这种鉴别方式破坏了大豆本质,并且存在检测费用高、效率低、精度差的问题。基于高光谱成像技术和机器学习理论,研究了大豆品种无损快速鉴别方法。采集并建立了4个品种(每个品种200粒,共计800粒)大豆的高光谱原始图像及光谱数据集。研究了利用归一化、均值中心化、小波变换、S-G平滑滤波以及矢量归一化对采集到的高光谱数据进行滤波去噪预处理,建立了基于KNN、RF及GBDT的大豆种粒无损检测模型。实验对比得出,利用主成分分析结合GBDT的检测模型精度最高,识别准确率可达99.58%。

The variety of soybean is directly related to the quality and oil yield of soybean products,and the detection of protein and fat content in soybeans is mainly used to achieve the identification of soybean varieties.This identification method destroys the essence of soybeans,and has the problems of high detection cost,low efficiency and poor accuracy.Based on hyperspectral imaging technology and machine learning theory,the non-destructive rapid identification method of soybean varieties was studied.Hyperspectral raw images and spectral datasets of 4 varieties(200 grains of each variety,a total of 800 grains)of soybeans were collected and established.The collected hyperspectral data were filtered and denoised by normalization,mean centralization,wavelet transform,S-G smoothing filtering and vector normalization,and a nondestructive testing model of soybean seed grain based on KNN,RF and GBDT was established.The experimental comparison indicated that the detection model using principal component analysis combined with GBDT had the highest accuracy,and the recognition accuracy can reach 99.58%.