文献类型：期刊文献

中文题名：基于多特征和改进BPNN的降香黄檀冠层叶片全氮含量无损诊断

英文题名：Nondestructive diagnosis of total nitrogen content in canopy leaves of Dalbergia odorifera based on multi-features and improved BPNN

作者：陈珠琳[1] 王雪峰[1] 陈毅青[2] 薛杨[2] 刘嘉政[1]

第一作者：陈珠琳

机构：[1]中国林业科学研究院资源信息研究所,北京100091;[2]海南省林业科学研究所,海口571100

年份：2019

卷号：38

期号：1

起止页码：275-282

中文期刊名：生态学杂志

外文期刊名：Chinese Journal of Ecology

收录：CSTPCD;;Scopus;北大核心:【北大核心2017】；CSCD:【CSCD2019_2020】；

基金：国家自然科学基金项目(31670642);林业科学技术推广项目([2016]11号)资助

语种：中文

中文关键词：营养诊断;全氮;PSO-BP-Adaboost;数字图像处理;降香黄檀

外文关键词：nutritional diagnosis;total nitrogen;PSO-BP-Adaboost;digital image processing;Dalbergia odorifera

分类号：S792.28

摘要：氮素是植物生命活动不可或缺的营养成分之一,合理的施肥不仅有利于植物健康成长,还可以减少对土壤及地下水污染等生态问题。本研究以降香黄檀为对象,提出了一种基于多特征和改进BPNN(Back Propagation Neural Network)的冠层叶片全氮含量无损诊断方法。通过数字图像处理技术对冠层图像进行分割,计算得到27种图像特征(颜色、纹理、形状),通过计算Pearson系数筛选出与全氮含量显著相关的因子,做主成分分析并提取前四主成分,将其作为改进BPNN(即PSO-BPNN-Adaboost)的输入向量。结果表明:使用多特征能更全面准确地反映降香黄檀冠层叶片的全氮含量;另一方面,BPNN、PSO-BPNN、BPNN-Adaboost与PSO-BPNN-Adaboost算法的比较结果表明,PSO-BPNN-Adaboost算法更可靠;同时,PSO处理对结果的优化程度更大,因此对于BP神经网络来说,需要先找一个合适的初始值和阈值,再对其进行增强处理。本研究考虑了氮胁迫对植株多方面的影响,突破了仅限于从颜色角度实现全氮含量预测的方法,也为珍贵树种经营中"精准施肥"提供了参考,可有效减小使用过量肥料造成的生态污染等问题。
Nitrogen is one of the indispensable nutritional components for plant growth. Proper fertilization is not only conducive to the healthy growth of plants,but also can reduce the pollution of soil and groundwater. In this study,we proposed a nondestructive diagnosis method of total nitrogen content in canopy leaves of Dalbergia odorifera based on multi-feature and improved BPNN. We divided the canopy image by digital image processing technology,and obtained 27 image features( color,texture,and shape). By calculating the Pearson coefficient,we screened out the factors significantly related to total nitrogen content by principal components analysis. We extracted the first four principal components as the input factors of improved BPNN which was optimized by PSO and Adaboost algorithm( PSO-BPNN-Adaboost). The results showed that the use of multiple features can more comprehensively and accurately reflect the nitrogen content of canopy leaves in Dalbergia odorifera. By comparing BPNN,PSO-BPNN,BPNN-Adaboost and PSOBPNN-Adaboost,we found that the PSO-BPNN-Adaboost was more reliable and the PSO algorithm had a more optimization effect. Therefore,for the BP neural network,it is necessary to find a suitable initial value and threshold before enhancing them. Our method fully considered the influence of nitrogen stress on plants and outcompeted the traditional method which only considered color factors. Our method provided reference for the precision fertilization in the management of precious trees,which could effectively reduce environmental pollution.