玉米籽粒霉变等级高光谱图像检测方法研究  

Study on hyperspectral image detection method of maize grain mildew grade

作　　者：杨东 王舒卉 吴建华 姜俊伊 宋凯[2] 石天玉 Yang Dong;Wang Shuhui;Wu Jianhua;Jiang Junyi;Song Kai;Shi Tianyu(Academy of National Food and Strategic Reserves Administration,National Engineering Research Center of Grain Storage and Logistics,Beijing 100037;Shenyang Ligong University,Shenyang 110159)

机构地区：[1]国家粮食和物资储备局科学研究院粮食储运国家工程研究中心,北京100037 [2]沈阳理工大学,沈阳110159

出　　处：《中国粮油学报》2022年第11期46-53,共8页Journal of the Chinese Cereals and Oils Association

基　　金：中央级公益性科研院所基本科研业务费专项资金(ZX1937)。

摘　　要：为了快速、无损检测出储藏玉米籽粒不同霉变状况,提升玉米收储环节质检效率,尝试利用高光谱成像技术结合机器学习算法构建玉米籽粒霉变等级分类模型。采集400~1 000 nm波段范围内玉米籽粒高光谱图像,以测定的真菌孢子数为依据,将籽粒霉变状态划分为健康、轻度霉变、中度霉变和重度霉变4个等级,采用随机蛙跳(RF)算法优选出7个光谱特征变量,针对特征波段图像,利用Tamura算法共提取出21个纹理特征变量,基于颜色矩阵提取出21个颜色特征变量。进一步结合支持向量机(SVM)、极限学习机(ELM)和偏最小二乘回归(PLSR)3种算法分别建立基于光谱、图像和图谱特征融合的玉米籽粒霉变等级分类模型。经分析比较,融合光谱和图像特征并结合ELM算法建立的分类模型用于玉米籽粒霉变等级识别效果最优,训练集和测试集分类准确率(Acc)分别为94.21%和93.86%,并将玉米籽粒霉变等级进行可视化表达。In order to quickly and undamaged detect the different mildew conditions of stored corn grains and improve the quality inspection efficiency of corn collection and storage, we tried to classification models of corn grain mildew by using hyperspectral imaging technology combined with machine learning algorithm in the present study. The hyperspectral images of corn grains in the range of 400~1 000 nm were collected. According to the measured number of fungal spores, the grain mildew state was divided into four grades: healthy, mild, moderate and severe mildew. Seven spectral characteristic variables were optimized by random frog(RF) algorithm. For the characteristic band images, 21 texture characteristic variables were extracted by Tamura algorithm, 21 color feature variables are extracted based on color moments. Further, combined with support vector machine(SVM), extreme learning machine(ELM) and partial least squares regression(PLSR), maize kernel mildew classification models based on spectral, image and spectral feature fusion was established respectively. Through analysis and comparison, the classification model based on spectral and image features combined with ELM algorithm could be used to identify corn grain mildew grade, and the classification accuracy(Acc) of training set and test set are 94.21% and 93.86% respectively.

关 键 词：玉米霉变籽粒 高光谱成像技术 随机蛙跳 极限学习机 

分 类 号：TS210.7[轻工技术与工程—粮食、油脂及植物蛋白工程]