基于计算机视觉和嗅觉的菠菜叶绿素含量检测方法研究  

Study on Spinach Chlorophyll Detection Method Using Computer Vision and Artificial Olfactory Sensor

作　　者：黄星奕[1] 徐海霞[1] 王顺[1] 潘思慧[1] 田潇瑜[1] 

机构地区：[1]江苏大学食品与生物工程学院,江苏镇江212013

出　　处：《现代食品科技》2017年第5期247-252,297,共7页Modern Food Science and Technology

基　　金：江苏省重点研发计划(现代农业)项目(BE2015308);江苏省高校自然科学研究重大项目(14KJA550001);江苏高校优势学科建设工程资助项目;国家自然基金项目(31671932);江苏省第四期"333工程"资助项目(BRA2015320)

摘　　要：以叶绿素含量为评价菠菜新鲜度的参考指标,开发菠菜采后品质无损检测方法。采用计算机视觉和电子鼻分别获取储藏期内菠菜的图像和气味信息。分别提取视觉、嗅觉信息的主成分作为模型的输入,以叶绿素含量的化学检测值作为模型的输出,采用误差反向传播神经网络建立菠菜叶绿素的定量预测模型。试验显示,以视觉信息为输入量的模型测试结果:训练集和测试集的均方根误差(RMSE)分别为0.1978 mg/g和0.2147 mg/g,相关系数(R)分别为0.8457和0.7995。以电子鼻信息为输入量的模型测试结果:训练、测试集的RMSE分别为0.3119 mg/g和0.3032 mg/g,R分别为0.7013和0.6905。以视觉和嗅觉融合信息为输入量的模型测试结果:训练、测试集的RMSE分别为0.1759 mg/g和0.2121 mg/g,R分别为0.8888和0.8736,精度比两个单一技术均有所提高。研究表明,利用计算机视觉和电子鼻技术预测菠菜叶绿素含量的方法是可行的,采用融合技术有助于提升模型的预测精度。Developing a rapid and nondestructive detection method for evaluating the quality of post-harvest spinach is essential for spinach producers and retailers. Typically, chlorophyll content was used as a reference index for spinach freshness, and the samples used in this study were stored at 4℃ for 12 d before being assessed by computer vision （for sample images） and electronic nose （for sample odors）. Feature variables extracted from image and odor information were analyzed by principal component analysis （PCA） method. A chlorophyll content prediction model was established with back propagation artificial neural network （BPANN）. The principal components （PCs） were used as the input parameters for the prediction model, and the chlorophyll content was used as the output. Experiments showed that the optimal prediction model of BPANN based on computer vision was obtained with four PCs. The root-mean-square error （RMSE） was 0.1978 mg/g and 0.2147 mg/g, and the correlation coefficient （R） was 0.8457 and 0.7995 for training and prediction sets, respectively. The results of the prediction model using BPANN based on the electronic nose showed that the RMSE was 0.3119 mg/g and 0.3032 mg/g, and R was 0.7013 and 0.7493 for training and prediction sets, respectively. The results of the BPANN model based on the fusion technique showed that the RMSE was 0.1759 mg/g and 0.2121 rag/g, while R was 0.8888 and 0.8736 for training and prediction sets, respectively. The prediction accuracy of the fusion technique was improved compaxed with that of either of the single detection methods. Overall, the results showed that it is feasible to predict the chlorophyll content of spinach during storage using computer vision and electronic nose, and the fusion technology is helpful in improving the prediction accuracy.

关 键 词：菠菜 计算机视觉 电子鼻 融合技术 人工神经网络 叶绿素 

分 类 号：S636.1[农业科学—蔬菜学] TP391.41[农业科学—园艺学]