摘要
氮素是植物生长需要的大量元素之一,生产经营者在植物生长过程中往往施加大量的氮肥,但过量的肥料会造成地下水污染.本文针对珍贵树种檀香,提出了一种基于ST-PCABP神经网络的檀香叶片全氮含量无损检测方法,为檀香的经营培育提供参考.结果表明:将野外获取到的檀香图像由RGB转换到L~*a~*b~*系统,可以较好地完成自然图像中的檀香分割.这是由于L~*a~*b~*系统色域宽,受光照变化的影响较小.ST-PCA-BP神经网络的特点是通过显著性检验(ST)筛选变量,使用方差膨胀因子和条件指数分析筛选结果的共线性,主成分分析法(PCA)消除共线性.该处理方法有效地减小了BP神经网络陷入局部最小值的概率,与原始数据、ST处理和PCA处理相比,决定系数最高,平均残差和均方根误差最小,是檀香叶片全氮含量无损、实时检测的最佳方法.
Nitrogen is one of the most important elements for plant growth. Producers often use a lot of nitrogen fertilizer during plant growth process. However,excessive fertilizer often cause groundwater pollution. In this study,we proposed a nondestructive testing method for total nitrogen content in leaves of sandalwood(Santalum album） based on ST-PCA-BP neural network. The results showed that,due to the wide color range of L＊a＊b＊color system and its robustness in illumination change,images obtained from the field which were converted from RGB to L＊a＊b＊color system had a satisfying segmentation result. The proposed ST-PCA-BP neural network was characterized by choosing variables through significance test(ST） and using variance inflation factor(VIF） and conditional index(CI） to analyze collinearity of results,and further using principal component analysis(PCA） to eliminate it. Such a method reduced the probability of the chance that BP neural network fell into the local minimum. Compared with the result of original data,data after ST processing,and data after PCA processing,the results of proposed method had the highest coefficient of determination,while the mean residual error and the root mean square error were the smallest. It was the best way to detect the total nitrogen content of sandalwood leaves with real-time and nondestructive method.
作者
陈珠琳
王雪峰
CHEN Zhu-lin, WANG Xue-feng(Institution of Forest Resources Information Technique, Chinese Academy of Forestry, Beijing 100091, China)
出处
《应用生态学报》
CAS
CSCD
北大核心
2018年第5期1551-1558,共8页
Chinese Journal of Applied Ecology
基金
国家自然科学基金项目(31670642)资助~~
关键词
檀香
BP神经网络
氮素
无损检测
Santalum album
BP neural network
nitrogen
nondestructive testing.
作者简介
陈珠琳，女，1994年生，硕士研究生．主要从事珍贵树种营养诊断研究．E-mail：825511059@qq．com;E-mail： xuefeng@ ifrit.ac.cn
