基于连续偏振光谱技术与嵌入型灰色神经网络的稻种发芽率检测方法研究

Study on Prediction of Rice Seed Germination Rate by Using Continuous Polarization Spectroscopy and Inlaid Grey Neural Network

针对稻种发芽率传统检测方法周期长,近红外光谱检测技术等无损检测方法受稻种自然颜色及含水量影响大的问题,通过连续偏振光谱结合嵌入型灰色神经网络(IGNN)的方法建立稻种发芽率预测模型。对检测连续偏振光谱运用经典模式分解(EMD)和小波包变换进行去噪处理,根据去噪效果选择EMD去噪。利用主成分分析(PCA)提取去噪后的连续偏振光谱特征,结合偏最小二乘法回归(PLSR)、反向传播神经网络(BPNN)、径向基神经网络(RBFNN)和IGNN分别构建稻种发芽率预测模型,建模结果显示10 min检测时间点IGNN预测模型精度最高,预测集相关系数RP=0.985,预测集均方根误差(RMSEP)为0.771。研究结果表明基于连续偏振光谱技术结合嵌入型灰色神经网络的方法实现稻种发芽率快速无损检测是可行的且精度较高。

With respect to the shortcomings of traditional rice seed germination rate detection such as more time consumption, and the problems of near infrared spectroscopy that it is easily influenced by natural color and water content of rice seeds, a method based on continuous polarization spectroscopy and inlaid grey neural network to achieve rapid and nondestructive prediction of rice seed germination rate is proposed. The obtained continuous polarization spectra are de-noised by the empirical mode decomposition（EMD） and wavelet packet transform,and EMD is selected according to the de-noising effect. The characteristics of de-noised continuous polarization spectra are extracted by the principal component analysis（PCA） and four modeling methods are used to build rice seed germination rate prediction models including partial least squares regression（PLSR）, back propagation neural network（BPNN）, radial basis function neural network（RBFNN） and inlaid grey neural network（IGNN）. The modeling results show that the IGNN model at 10 min testing time is the most accurate, with the correlation coefficient of prediction set as 0.985 and mean square error of prediction set as 0.771. The research results show that the method based on continuous polarization spectroscopy and inlaid grey neural network can achieve rapid and nondestructive prediction of rice seed germination rate and has high prediction accuracy.