高光谱结合优化算法的多品种高粱混合的淀粉含量检测 |
| |
引用本文: | 补友华,姜鑫娜,田建平,胡新军,黄浩平,高剑,黄丹,罗惠波. 高光谱结合优化算法的多品种高粱混合的淀粉含量检测[J]. 中国粮油学报, 2022, 37(11): 236-244 |
| |
作者姓名: | 补友华 姜鑫娜 田建平 胡新军 黄浩平 高剑 黄丹 罗惠波 |
| |
作者单位: | 四川轻化工大学机械工程学院,四川轻化工大学机械工程学院,四川轻化工大学机械工程学院,四川轻化工大学机械工程学院,四川轻化工大学机械工程学院,四川轻化工大学机械工程学院,四川轻化工大学生物工程学院,四川轻化工大学生物工程学院 |
| |
基金项目: | 四川省科技厅重点研发项目(2019YFG0167),四川轻化工大学研究生创新基金项目(y2021025) |
| |
摘 要: | 高粱作为一种酿酒原料,其不同混合比例配比的高粱的淀粉含量会影响白酒的品质和产量。因此,准确高效地检测混合高粱中的淀粉含量对获得优质高产的白酒具有重要意义。本文基于可见光高光谱成像技术研究了混合高粱中的淀粉含量的快速检测方法。采用不同预处理方法对高粱样本的光谱数据进行预处理,并建立偏最小二乘法回归(PLSR)模型来确定最佳预处理方式。使用主成分分析(PCA)、PLSR算法分别提取高粱样本的光谱特征。基于全波长和光谱特征分别建立了预测高粱淀粉含量的遗传算法-BP神经网络(GA-BPNN)和粒子群算法-支持向量机回归(PSO-SVR)模型。对比模型性能发现,采用PCA方法提取的光谱特征建立的GA-BPNN模型最优,其直链淀粉的预测决定系数、预测均方根误差分别为0.992 2、0.041 6,支链淀粉的预测决定系数、预测均方根误差分别为0.933 6,0.151 9。研究结果表明,可见光高光谱成像技术结合优化算法可以快速预测不同混合比例配比下高粱的淀粉含量,为检测高粱的淀粉含量提供了一种新的方法。
|
关 键 词: | 高光谱成像 高粱淀粉含量 主成分分析 偏最小二乘法 BP神经网络 |
收稿时间: | 2021-11-19 |
修稿时间: | 2022-02-17 |
Detection of starch content in multi-species sorghum mixture by hyperspectral combined with optimization algorithm |
| |
Abstract: | Sorghum is a raw material for brewing, and its starch content in different mixture ratios affects the quality and yield of liquor. Therefore, accurate and efficient detection of starch content in mixed sorghum is of great significance for obtaining high-quality and high-yield liquor. In this paper, a rapid detection method of starch content in mixed sorghum was studied based on visible light hyperspectral imaging technology. Different pre-processing methods were used to pre-process the spectral data of sorghum samples and a partial least squares regression (PLSR) model was developed to determine the best pre-processing method. The spectral features of sorghum samples were extracted using principal component analysis (PCA) and PLSR algorithms, respectively. Based on the full wavelength and spectral features, genetic algorithm-back propagation neural network (GA-BPNN) and particle swarm algorithm-support vector machine regression (PSO-SVR) models were developed to predict the starch content of sorghum, respectively. Compared with the model performance, GA-BPNN model based on spectral features extracted by PCA method was the best, and the predictive decision coefficient and root mean square error of amylose were 0.992 2 and 0.041 6, respectively, and those of amylopectin were 0.151 9 and 0.933 6, respectively. The results show that visible hyperspectral imaging combined with optimized algorithms can rapidly predict the starch content of sorghum at different mix ratio ratios, providing a new method for detecting the starch content of sorghum. |
| |
Keywords: | hyperspectral imaging sorghum starch content principal component analysis partial least squares method BP Neural Network |
|
| 点击此处可从《中国粮油学报》浏览原始摘要信息 |
|
点击此处可从《中国粮油学报》下载全文 |
|