基于PCA-1DCNN的近红外光谱粮食作物主要成分检测方法

针对传统的近红外光谱定量技术难以选择合适的光谱预处理方法且模型预测精度低的问题，以3个谷物数据集的近红外光谱数据集为研究对象，构建了基于主成分分析光谱筛选算法的一维卷积神经网络模型。与传统的偏最小二乘回归和支持向量机模型的性能做了对比后，一维卷积神经网络构建的模型性能均为最优。其中在对玉米数据集的水分、油脂、蛋白质、淀粉的定量建模中，模型的决定系数分别为99.09%、98.15%、98.89%、99.60%；在对grain数据集的定量建模中，四种成分模型的决定系数分别为100%、100%、100%、99.99%；在对小麦数据集的定量建模中，小麦蛋白质模型的决定系数为99.80%。为了验证主成分分析光谱筛选算法对粮食作物主要成分定量回归模型的有效性，在3个光谱数据集上去除了主成分分析算法进行消融实验。研究结果表明：基于主成分分析算法与一维卷积神经网络的回归建模方法为粮食作物成分含量的检测提供一种快速无损精确的判定方式，研究结果对于粮食作物成分的含量检测具有促进作用。

Detection Method of Main Components of Food Crops by Near Infrared Spectroscopy Based on PCA and 1DCNN

Aiming at the problem that the traditional near-infrared spectroscopy quantitative technology is difficult to select the appropriate spectral preprocessing method and the low prediction accuracy of the model, taking the near-infrared spectroscopy data sets of 3 grain datasets as the research object, a one-dimensional convolutional neural network model based on principal component analysis spectral screening algorithm was constructed. Compared with the traditional partial least squares regression and support vector machine model, the performance of the model constructed by one-dimensional convolutional neural network is the best. In the quantitative modeling of water, oil, protein and starch in corn data set, the determination coefficients of the model are 99.09%, 98.15%, 98.89% and 99.60%. In the quantitative modeling of grain data set, the determination coefficients of the four component models are 100%, 100%, 100% and 99.99%. In the quantitative modeling of wheat data set, the determination coefficient of wheat protein model is 99.80%. In order to verify the effectiveness of the principal component analysis spectral screening algorithm on the quantitative regression model of the main components of food crops, the ablation experiment was carried out by removing the principal component analysis algorithm in the three spectral datasets.The results show that the regression modeling method based on principal component analysis algorithm and one-dimensional convolutional neural network provides a fast and lossless and accurate determination method for the detection of food crop component content, and the research results have a promoting effect on the content detection of food crop components.