基于高光谱成像技术的糯玉米种子分类研究
Classification of Waxy Corn Seeds Varieties based on Hyperspectral Imaging Technology
庄浩轩 1魏明生 1王波 2赵慕阶 3陈化东3
作者信息
- 1. 江苏师范大学物理与电子工程学院 江苏,徐州 221116
- 2. 徐州市检验检测中心 江苏,徐州 221111
- 3. 江苏艾龙科技有限公司 江苏,南京 211112
- 折叠
摘要
为了快速、准确、无损地对糯玉米种子分类,采用可见-近红外(400～1000 nm)高光谱成像仪对5种糯玉米种子进行数据采集,使用一阶中心差分联合SG平滑对糯玉米种子的原始光谱数据进行预处理去噪,通过自优化竞争性自适应重加权采样算法筛选出56个重要的特征波段,同时采用灰度共生矩阵和Sobel算子提取糯玉米种子的相关性、能量、同致性、相关熵、灰度熵和梯度熵等6种纹理特征,将光谱特征与纹理特征融合后构建支持向量机分类模型,分别用350个训练样本、150个测试样本和50个预测样本对模型进行训练、测试和预测分类,相应得到了准确率为98.50％、95.92％和94.00％的最佳结果,表明利用高光谱成像技术对糯玉米种子分类是可行的.
Abstract
In order to quickly,accurately,and losslessly classify waxy corn seeds,a visible near-infrared(400～1000 nm)hyperspectral imager was used to collect data from 5 types of waxy corn seeds.The original spectral data of glutinous corn seeds were preprocessed using 1st central difference combined with SG smoothing.A automatic optimizing competitive adaptive reweighted sampling algorithm was used to screen 56 important feature bands,At the same time,the gray level co-occurrence matrix and Sobel operator were used to extract six texture features of waxy corn seeds,including correlation,energy,congruence,correlation entropy,gray level entropy,and gradient entropy.After fusing spectral features with texture features,a support vector machine classification model was established.The best results were obtained for the training set of 98.50％,the test set of 95.92％,and the prediction set of 94％.It is feasible to use hyperspectral imaging technology to classify glutinous corn seeds.
关键词
高光谱成像技术/一阶中心差分/自优化/竞争性自适应重加权采样算法/灰度共生矩阵Key words
hyperspectral imaging technology/1st central difference/automatic optimization/competitive adaptive reweighted sampling algorithm/gray level co-occurrence matrix引用本文复制引用
基金项目
南京市科技发展计划项目(2021SX00000519)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据