基于图像处理技术的小麦水分检测方法研究  

Research on Detection Method of Wheat Moisture Based on Image Processing Technology

作　　者：张晋宁 尹君 田一骏 Zhang Jinning;Yin Jun;Tian Yijun(Academy of National Food and Strategic Reserves Administration,Beijing 100037;College of Information Science and Engineering,Henan University of Technology,Zhengzhou 450000)

机构地区：[1]国家粮食和物资储备局科学研究院,北京100037 [2]河南工业大学信息科学与工程学院,郑州450000

出　　处：《中国粮油学报》2023年第10期211-215,共5页Journal of the Chinese Cereals and Oils Association

基　　金：国家粮食和物资储备局科学研究院自选课题(JY2202)。

摘　　要：水分是小麦安全储藏的重要指标之一。利用图像处理技术对采集到的不同水分的小麦图像进行特征提取,提取小麦籽粒图像中的47个图像特征(23个形态特征和24个颜色特征),利用SPSS软件对其进行相关性分析。分析图像特征与小麦含水量的相关性,使用逐步选择法筛选出对于小麦水分预测最重要的特征子集并建立了ElasticNet、RandomForest、AdaBoost回归模型。结果表明,3个回归模型的R 2值分别为0.96、0.97、0.98,RMSE值分别为1.05、0.26、0.84。其中RandomForest模型预测结果的水分绝对误差主要分布在±1.5之间。因此,可基于图像处理技术用于小麦水分无损检测的研究。Moisture is one of the important indicators for safe storage of wheat.The features of the collected wheat images with different moisture content were extracted by using image processing technology,47 image features(23 morphological features and 24 color features)in wheat grain images were extracted,and the SPSS software was used to conduct the correlation analysis on them.The correlation between image features and wheat moisture was analyzed,the feature subsets that are most important for the predication of wheat moisture were screened out by using the stepwise selection,and ElasticNet,RandomForest,AdaBoost regression models were established.The results showed that the R 2 values of the three regression models were 0.96,0.97 and 0.98,respectively,and the RMSE values were 1.05,0.26 and 0.84,respectively.Among them,the absolute error of moisture predicted by RandomForest model is mainly distributed between±1.5.Therefore,the nondestructive testing of wheat moisture could be researched based on the image processing technology.

关 键 词：图像处理 小麦 水分检测 

分 类 号：S512.1[农业科学—作物学] TP391.4[自动化与计算机技术—计算机应用技术]