基于导数光谱与主成分分析的小麦籽粒赤霉病识别

Recognition of wheat grain scab based on derivative spectrum and principal component analysis

赤霉病是威胁我国食品安全的主要病害之一,为了给粮食加工和作物育种提供技术支持,迫切需要探索小麦籽粒赤霉病的识别方法.利用高光谱成像仪扫描健康和感染赤霉病的小麦籽粒获取高光谱图像.使用图像处理技术分离籽粒和背景后,用多元散射校正、1阶导数和2阶导数对光谱数据进行预处理.利用主成分分析提取小麦籽粒的光谱特征.使用不同模型对籽粒进行识别,比较多种处理组合的评估指标.结果表明:1阶导数—主成分分析—BP神经网络组合的总体分类精度最高,其值为91.67%.

Fusarium head blight is one of the main diseases threatening China’s food safety.In order to provide technical support for grain processing and crop breeding,it is urgent to explore the rapid detection method of wheat grain scab.Hyperspectral imager was used to scan healthy and fusarium-infected wheat grain to obtain high-spectral images.After separating the grain and background based on image processing technology,the spectral data was processed by multivariate scattering correction(MSC)technique,first derivative spectrometry and second derivative spectrometry.Principal component analysis(PCA)was used to extract the features of wheat grain.Different models were used to recognize wheat grains.And the evaluation indicators of a combination of different process methods were compared with each other.The results showed that the combination of first derivative spectrometry,PCA and back propagation neural network(BPNN)had the highest overall accuracy which was 91.67%.