光谱学与光谱分析, 2018, 38 (8): 2516, 网络出版: 2018-08-26  

基于可见光波段的色彩概率聚类模型的玉米杂交种子识别

Hybrid Seed Recognition of Maize Based on Probability Clustering Model Using Visible Light Color Features
作者单位
1 山东农业大学机械与电子工程学院, 山东 泰安 271018
2 山东省园艺机械与装备重点实验室, 山东 泰安 271018
3 山东农业大学农学院, 山东 泰安 271018
摘要
由于同一种玉米杂交种子籽粒的粒型多样、 色彩随贮藏时间不同而不同, 仅靠其形状和单一区域色彩的机器视觉方法识别种类较为困难, 且现有识别算法多以高光谱特征作为分类基础, 对于不同时期、 不同种类的玉米杂交种子要分别训练分类识别器, 识别前需要大量的训练工作。 为提高玉米种子品种识别方法的适用性, 根据花粉直感色彩遗传现象, 提出以可见光波段玉米种子的多区域小波色彩特征作为识别参数, 建立多模型的概率聚类模型进行玉米杂交种子种类识别。 该方法首先采用专有设备采集单粒玉米种子的无胚芽侧和顶端两部分色彩信息, 包括RGB, HIS和Lab色彩信息, 对该色彩信息进行增强和特征优化选择, 通过小波包分解提取优化出21维细节识别向量; 其次采用不同聚类模型对优化后色彩特征进行聚类识别, 建立基于SOM、 K-means、 两步法三种聚类识别模型; 最后以多种聚类模型结果为基础, 建立基于概率模型的玉米种子品种识别。 通过对郑单958、 先玉335、 郑58(郑单958母本)、 昌7-2(郑单958父本)、 PH6WC(先玉335母本)、 PH4CV(先玉335父本)的试验, 发现该方法可有效识别非亲缘关系和父本亲缘关系的玉米种子, 识别率可达到98%以上; 而对于亲缘关系母本识别率可达到75%。 采用可见光波段玉米种子多区域色彩特征, 结合概率聚类模型的方法可为玉米杂交种子纯度在线检验识别提供科学依据。
Abstract
Because the grain size of the same kind of hybrid maize seed is different and the maize color changes as the storage time varies, it is difficult to identify the species only by the machine vision method with its shape and color in a single region. Besides, the existing recognition algorithm mostly uses the hyper-spectral feature as the basis for classification, so for different periods, different types of hybrid maize seeds need to be trained by the classification equipment, and a lot of training is required before the identification. In order to improve the applicability of identification method for maize seed variety, a multi-model probabilistic clustering method was established based on the multi-regional wavelet color characteristics of maize seed in the visible light band as the recognition parameter. This method used a special equipment to extract the non-germinal and the topside color information of the single-grain maize seed, including the color information of RGB, HIS and Lab. Then the color information was enhanced, the feature selection was optimized and the 21-dimensional detail recognition vector was perfected by wavelet packet decomposition. Secondly, the clustering recognition of the optimized color feature was carried out by different clustering models. Three clustering models based on SOM, K-means and two-step method were thus established. Finally, based on the results of multiple clustering models, the maize seed variety identification via probability model was set up. Through the experiments on Zheng Dan 958, Xian Yu 335, Zheng 58 (Zheng Dan 958 female), Chang 7-2 (Zheng Dan 958 male), PH6WC (Xian Yu 335 female), PH4CV (Xian Yu 335 male), it was shown that the method was able to effectively identify maize seeds with non-genetic relationship and parental relationship, with the recognition rate reaching over 98%; While the recognition rate of female parent was 75%. This can provide scientific basis for on-line identification of hybrid seed purity. The method of probability clustering model can provide scientific basis for the identification of maize hybrid seed purity by using visible light multi-regional color characteristics.

刘双喜, 张宏建, 王金星, 王震, 张春庆, 李岩. 基于可见光波段的色彩概率聚类模型的玉米杂交种子识别[J]. 光谱学与光谱分析, 2018, 38(8): 2516. LIU Shuang-xi, ZHANG Hong-jian, WANG Jin-xing, WANG Zhen, ZHANG Chun-qing, LI Yan. Hybrid Seed Recognition of Maize Based on Probability Clustering Model Using Visible Light Color Features[J]. Spectroscopy and Spectral Analysis, 2018, 38(8): 2516.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!