激光与光电子学进展, 2016, 53 (12): 123001, 网络出版: 2016-12-14
基于高光谱成像技术与CARS算法的玉米种子含水率检测 下载: 606次
Moisture Content Detection of Maize Kernels Based on Hyperspectral Imaging Technology and CARS
光谱学 高光谱检测技术 竞争性自适应重加权变量选择算法 玉米种子 正反面 含水率 spectroscopy hyperspectral detection technology competitive adaptive reweighted sampling algorithm maize kernel front and reverse side moisture content
摘要
为实现玉米种子含水率(MC)的精确、快速、无损检测, 消除种子放置方式(胚部朝上/下)的影响, 基于高光谱成像和图像处理技术, 结合变量筛选法, 针对玉米种子正反面放置的不同分别建立对应的MC预测模型。分别采集种子正、反两面高光谱图像, 提取质心区域光谱数据, 采用竞争性自适应重加权变量选择算法筛选特征波段, 建立对应的MC预测模型。对比图像不同部位光谱曲线变化趋势, 挑选4个特征波段(1104, 1304, 1454, 1751 nm)进行波段运算获取种子正、反面信息及质心位置。依据正、反面检测结果, 自主选择对应的MC预测模型对45个验证集样本进行含水率检测。结果表明, 使用波段运算正、反面识别率分别为97.8%、100%; 正、反两面验证集相关系数分别为0.969, 0.946, 均方根误差分别为0.464%, 0.616%。该研究为使用多光谱成像技术实现玉米种子MC的快速无损自动化检测奠定基础。
Abstract
In order to realize accurate, rapid and nondestructive detection for moisture content (MC) of maize kernel and avoid effects of placement state (embryo up or down) on detection results, a novel detection method based on hyperspectral imaging and image processing techniques is proposed. The variable selection method is used to establish MC prediction model according to the placement state of maize kernel. The hyperspectral images including both front and reverse side of maize kernel are acquired, spectral data in centroid region is extracted, and competitive adaptive reweighted sampling algorithm is used for characteristic wavelength selection. And the prediction models including front and reverse side prediction model are built for MC prediction. The spectral curves in different parts of hyperspectral images are contrasted mutually to judge if the maize kernel appeared in the image is front side upward (embryo up) or not, and four wavebands (1104, 1304, 1454, 1751 nm) are selected for front and reverse side detection with band math. The MC of 45 validation set samples are detected with the proposed algorithm. Results show that the accuracy of front and reverse side detection is about 97.8%, 100%, respectively, the validation set correlation coefficient of front and reverse side are 0.969, 0.946, respectively, the root mean square error are 0.464%, 0.616%, respectively. This research establishes foundation for the MC detection of maize kernel with multi-spectral technique.
王超鹏, 黄文倩, 樊书祥, 张保华, 刘宸, 王晓彬, 陈立平. 基于高光谱成像技术与CARS算法的玉米种子含水率检测[J]. 激光与光电子学进展, 2016, 53(12): 123001. Wang Chaopeng, Huang Wenqian, Fan Shuxiang, Zhang Baohua, Liu Chen, Wang Xiaobin, Chen Liping. Moisture Content Detection of Maize Kernels Based on Hyperspectral Imaging Technology and CARS[J]. Laser & Optoelectronics Progress, 2016, 53(12): 123001.