基于高光谱成像技术的冬瓜种子品种鉴别  

Identification of Winter Melon Seed Varieties Based on Hyperspectral Technology

作　　者：刘文俊 张淑娟[1] 张立秀 王润润 LIU Wenjun;ZHANG Shujuan;ZHANG Lixiu;WANG Runrun(College of Agricultural Engineering,Shanxi Agricultural University,Jinzhong Shanxi 030801,China)

机构地区：[1]山西农业大学农业工程学院,山西晋中030801

出　　处：《农业工程》2022年第12期61-65,共5页AGRICULTURAL ENGINEERING

基　　金：山西省重点研发计划项目(201903D221027);山西省研究生创新项目(2021Y329)。

摘　　要：为实现冬瓜种子品种的快速无损检测,利用高光谱成像技术采集了广东黑皮、一串铃和韩育粉皮3个品种冬瓜种子的光谱图像信息。采用标准归一化(SNV)、多元散射校正(MSC)、中值滤波(MF)等方法进行预处理,利用连续投影法(SPA)、回归系数法(RC)和竞争性自适应算法(CARS)提取特征波长,建立了粒子群优化BP神经网络判别模型,预测集的决定系数与均方根误差分别为R2=0.930、RMSEP=0.047,准确率高达96.30%。研究表明,高光谱技术结合粒子群优化BP神经网络可以实现对冬瓜种子品种快速无损鉴别。In order to achieve rapid non-destructive testing of winter melon seed varieties,spectral image information of winter melon seeds of three varieties of Guangdongheipi,Yilchuanling and Hanyufenpi was collected by hyperspectral imaging technology.Standard normalization(SNV),multiple scattering correction(MSC),median filtering(MF)and other methods were used for pretreatment.Feature wavelengths were extracted using successive projection algorithm(SPA),regression coefficient(RC),and competitive adaptive algorithm(CARS).Particle swarm optimization BP neural network discriminant model was established,and coefficient of determination and root mean square error of prediction set were R2=0.930 and RMSEP=0.047,respectively,and accuracy rate was as high as 96.30%.Results showed that hyperspectral technology combined with particle swarm optimization BP neural network could achieve rapid and non-destructive identification of winter melon seed varieties.

关 键 词：高光谱 品种鉴别 粒子群 BP神经网络 冬瓜种子 

分 类 号：S123[农业科学—农业基础科学]