摘要
【目的】檀香是一种典型珍贵树种,在幼龄期时,不合理的田间施肥会影响其正常生长,降低存活率。因此,本文提出了一种基于可见光-近红外图像的幼龄檀香全磷营养诊断方法,为实时监测珍贵树种生长状态及养分需求提供参考。【方法】通过将野外获取的檀香图像转换到HSI颜色空间,提取S和I通道,利用二者在使用Otsu分割后产生的优势互补,并结合形态学运算,从复杂背景中提取出檀香。计算出形状、纹理和光谱及植被指数特征后,分别使用显著性分析(ST)和平均影响值(MIV)方法进行变量筛选,并使用遗传算法(GA)初始化BP神经网络的权值和阈值,最终得到预测结果。【结果】(1)复杂背景下的檀香分割中,S通道和I通道相结合可以将大部分背景(天空、土壤、其他绿色植物)与目标檀香分割开,同时结合7×7中值滤波、形态学运算和超G因子,将其他毛刺去除。与常用的支持向量机相比,本文提出的分割算法结果更接近于目视解译,像素数和颜色误差更小。(2)对不同施磷水平下各特征进行分析发现,适当增加施磷量有利于促进叶绿素的形成,使得纹理更均匀清晰,加快叶片生长;当过量时则会破坏叶绿体,造成叶片组织出现变化,导致叶片黄化,叶片出现网状脉纹,增加了纹理复杂程度。(3) ST与MIV筛选出的变量差异较大,通过GA-BPNN训练结果可知,MIV方法筛选出的变量对全磷含量的影响更大,预测集得到的决定系数达到0. 801,平均残差为0. 032 g/kg,均方根误差为0. 666 g/kg。【结论】通过处理可见光-近红外图像,实现了幼龄檀香的全磷含量诊断,有效提高了磷肥利用率,同时也可以减小过量施肥引起的地下水污染等生态问题。
[Objective]Sandalwood is a typical precious tree species. During its young stage,more or less fertilization will affect its growth and reduce survival rate. In this paper,a total phosphorus nutrition diagnosis method for young sandalwood based on visible light and near infrared image recognition is proposed. It provides a reference for real-time monitoring of the growth state and nutrient requirements of precious tree species. [Method]S and I channels were extracted after converting field acquired sandalwood images to HSI color space. By combing the advantage of S and I channels segmentationresults using Otsu method and morphological operation,sandalwood was extracted from the complex background. We used different methods to optimize BP neural network. On one hand,ST and MIV methods were used to select the variables in shape,texture,spectrum and vegetation index. On the other hand,genetic algorithm( GA) was used to initialize the weights and thresholds,and the prediction results were finally obtained. [Results]( 1) In the complex background of sandalwood segmentation,the combination of H channel and S channel successfully separated most of the background( sky,soil and other green plants) from target sandalwood. Median filter sized in 7 × 7,morphological operation and super G factor were used to remove other burrs.( 2) The characteristics under different levels of phosphorus application showed that the appropriate increase fertilizer could promote the compound of chlorophyll,make the texture more uniform and clear,and increase the growth of the leaves. When the application exceeded the best value,the chloroplast could be destroyed,texture changed and leaf color turned into yellow.( 3) The variables selected by ST and MIV were different. GA-BPNN training results showed that the variables selected by the MIV method had greater influence on the total phosphorus content. The determinant coefficient of the prediction set was 0. 801,the mean residual was 0. 032 g/kg,and the root mean square error was 0. 666 g/kg. [Conclusion] In this paper,the total phosphorus content of young sandalwood was predicted by processing the visible light and near infrared image. By this method,the utilization rate of phosphate fertilizer will be improved effectively,and the ecological problems such as groundwater pollution caused by excessive fertilization could also be reduced.
作者
陈珠琳
王雪峰
孙汉中
Chen Zhulin;Wang Xuefeng;Sun Hanzhong(Research Institute of Resource Information Technique,Chinese Academy of Forestry,Beijing 100091,China;Inner Mongolia Deerbuer Forestry Bureau,Genhe 022357,Inner Mongolia,China)
出处
《北京林业大学学报》
CAS
CSCD
北大核心
2019年第2期88-96,共9页
Journal of Beijing Forestry University
基金
中央级科研院所基本科研业务费专项项目(CAFYBB2014MA006)
林业科学技术推广项目([2016]11号)
关键词
檀香
可见光
近红外
全磷含量
图像分割
sandalwood
visible light
near infrared band
total phosphorus content
image segmentation
