林业科学  2018, Vol. 54 Issue (5): 127-134 PDF
DOI: 10.11707/j.1001-7488.20180514
0

#### 文章信息

Xiao Xiayang, Wen Jian, Xiao Zhongliang, Li Weilin, Zhang Houjiang

Detection and Recognition of Tree Trunk Internal Structure Based on Radar

Scientia Silvae Sinicae, 2018, 54(5): 127-134.
DOI: 10.11707/j.1001-7488.20180514

### 作者相关文章

Detection and Recognition of Tree Trunk Internal Structure Based on Radar
Xiao Xiayang, Wen Jian , Xiao Zhongliang, Li Weilin, Zhang Houjiang
School of Technology, Beijing Forestry University Beijing 100083
Abstract: 【Objective】 In order to provide a theoretical basis for the analysis, localization and distribution characterization of internal structure and defects of the radar waves by non-destructive testing technique, tree radar tomography were applied to detect internal defects of wood body, and the recognition algorithm of internal reflection features of radar trees was studied and analyzed. 【Method】 Tomography of Willow wood was carried out with a 900 MHz medium-coupled tree radar. The threshold value, matched filter and Hilbert's algorithm were used to obtain the reflection position of the defect layer. Based on the layer stripping inversion method, the inversion of the dielectric constant were applied for the different media in the tree, and the relative depth of the horizon were calculated, then the point cloud data of the contour of the tree obtained by the 3D laser scanner were mapped, and the absolute position of the B-scan radar image and the contour of the fault were mapped according to the contour tracing method. Therefore, the accurate positioning and characterization of the internal defects of trees were achieved. The three algorithms were respectively verified by finite difference-time-domain (FDTD) forward method, and the method is applied to the test of the ancient willow wood specimen in the Summer Palace.【Result】 The results of forward comparison test show that Hilbert plot method is better than threshold method and matched filter method in identifying defects in wood body. The actual willow test results show that the depth error of defects in trees is 10%. The area error of defects calculated by 3D laser and contour scanning technology is about 5%.【Conclusion】 The method proposed in this paper can realize the accurate localization and imaging distribution of the internal defects of the trees by the radar wave scanning images.
Key words: radar wave    laser scanning    threshold    match filter    Hilbert    locate abnormal

1 材料与方法 1.1 雷达波扫描图像获取

 图 1 数据采集设备 Figure 1 Devices of data acquisition

 图 2 数据采集 Figure 2 Data acquisition
1.2 外形轮廓获取

1.3 雷达波原理

 图 3 正演模型 Figure 3 The first forward model

 图 4 单道回波信号 Figure 4 The single channel of reflective signal

 ${y_{\rm{r}}}\left( t \right) = \sum\limits_{i = 0}^N {{A_i}x\left( {t - {\tau _i}} \right)} + n\left( t \right)。$ (1)

 ${v_{\rm{p}}} = c/\sqrt {\varepsilon '} ;$ (2)

 $Z = \frac{{c \times t}}{{2 \times \sqrt {\varepsilon '} }}。$ (3)

 ${\varepsilon _1} = {\left( {\frac{{1 + {A_1}/{A_{\rm{m}}}}}{{1 - {A_1}/{A_{\rm{m}}}}}} \right)^2}。$ (4)

 ${\varepsilon _2} = {\varepsilon _1}{\left[ {\frac{{1 - {{\left( {{A_1}/{A_{\rm{m}}}} \right)}^2} + \left( {{A_2}/{A_{\rm{m}}}} \right)}}{{1 - {{\left( {{A_1}/{A_{\rm{m}}}} \right)}^2} - \left( {{A_2}/{A_{\rm{m}}}} \right)}}} \right]^2}。$ (5)

1.4 雷达波对树木分层处理方法

1.4.1 阈值法

 ${V_{\rm{t}}} = \delta \sqrt { - 2\lg {P_{\rm{f}}}} 。$ (6)

GPR信号中的回波脉冲通常是多峰的，单个反射脉冲有多个局部峰值点，对缺陷点定位分层产生了干扰。信号包络可以解决多重峰的问题，包络信号通常是在通信系统中使用以提取调制包络和拒绝高频载波信号的。本文使用的GPR信号虽然属于无载波脉冲体制，但是依然可以提取出信号幅度变化趋势的包络：

 ${x_{\rm{e}}}\left( t \right) = \left| {{x_{\rm{a}}}\left( t \right)} \right| = \left| {x\left( t \right) + j\mathop {x\left( t \right)}\limits^ \wedge } \right|。$ (7)

 图 5 阈值法分层 Figure 5 The layered graphic by threshold
1.4.2 匹配滤波器法

 $h\left( t \right) = x\left( {T - t} \right)。$ (8)

 ${S_{\rm{t}}} = {\rm{erf}}{{\rm{c}}^{ - 1}}\left( {{P_{\rm{f}}}} \right)\sqrt {{\sigma ^2}E} 。$ (9)

 图 6 匹配滤波器法分层 Figure 6 The layered graphic by matched filter
1.4.3 希尔伯特积算法

Huang等(2008)提出了一种新的时频分析方法——希尔伯特-黄变换(Hilbert-Huang transfrom，HHT)，该方法对于非平稳、非线性信号的分析比较直观，且自适应强。在这个理论中，通过经验模态分解(empirical mode decomposition，EMD)将信号自适应分解成有限多个内在模分量(intrinsic mode function，IMF)和1个表征信号趋势变化的残余信号，并且提出对得到的各个IMF运用希尔伯特变换进行时频分析。

1) 对回波信号进行EMD分解，得到相应的N个IMF分量c1(t)~cN(t)。

2) 计算所有IMF分量绝对值的点积，记为：

 $P\left( t \right),P\left( t \right) = \left| {\prod\limits_{i = 1}^N {{c_i}\left( t \right)} } \right|。$ (10)

3) 因为回波信号包含正峰和负峰，使得接收信号由多峰脉冲组成，易造成虚假检测，因此需要对信号进行窗函数的平滑处理。用P(t)和窗函数W(t)进行卷积来完成平滑处理：

 $T\left( t \right) = W\left( t \right) * P\left( t \right)。$ (11)

 图 7 希尔伯特积算法分层 Figure 7 The layered graphic by Hilbert
2 结果与分析 2.1 分层处理算法比较

 ${\rm{error\% = }}\frac{{{\rm{d}}y - {\rm{d}}x}}{{{\rm{d}}y}} \times 100\% 。$ (12)
 图 8 楔形正演模型 Figure 8 Wedge forward model

 图 9 3种算法的误差率 Figure 9 The error rate of the three algorithms
2.2 试验数据分析

 图 10 试件T1、T2和T3相关分析 Figure 10 Analysis figure of Sample T1, T2 and T3

B扫描图通过雷达天线在树木某一高度进行360°环绕探测获取，实际树木缺陷面积通过网格法计算，试验结果与Treewin分析结果进行对比分析。Treewin采用单一介电常数结合阈值法对雷达回波数据进行定位，结合极坐标转换生成树木内部缺陷的二维图像(图 10c)，通常与实际树木异相差较大，缺陷区域面积误差率在20%左右。本研究采用希尔伯特积算法对树木进行分层，层剥反演介电常数实现树木内缺陷位置定位，结合三维激光对树木外形轮廓的数据，利用轮廓追踪和坐标变换方法生成不规则树木的二维缺陷分层图(图 10d)。令Sj为检测出的腐朽缺陷面积，Sz为实际缺陷的面积，通过对比试件图像或检测图像，用网格法来确定SjSz。可得面积误差率S%为：

 $S{\rm{\% = }}\frac{{\left| {{S_{\rm{j}}} - {S_{\rm{z}}}} \right|}}{{{S_{\rm{z}}}}} \times 100\% 。$ (13)

3 结论