Journal of Forestry Research, 18(3): 221−225 (2007) 221
DOI: 10.1007/s11676-007-0045-5
Effect of sensor quantity on measurement accuracy of log inner defects
by using stress wave
WANG Li-hai, XU Hua-dong, ZHOU Ci-lin, LI Li, YANG Xue-chun
Northeast Forestry University, Harbin 150040, P. R. China
Abstract: Wood nondestructive testing (NDT) is one of the high efficient methods in utilizing wood. This paper explained the principle of
log defect testing by using stress wave, and analyzed the effects of sensor quantity on defect testing results by using stress wave in terms of
image fitting degree and error rate. The results showed that for logs with diameter ranging from 20 to 40 cm, at least 12 sensors were needed
to meet the requirement which ensure a high testing accuracy of roughly 90% of fitness with 0.1 of error rate. And 10 sensors were recom-
mended to judge the possible locations of defects and 6 sensors were sufficient to decide whether there were defects or not.
Keywords: Sensor quantity; Log defect testing; Stress wave; Image fitting degree
Introduction
Quality assessment and defects detection of woods had the im-
portant effects on forestry industry. Nondestructive evaluation on
the properties of logs is necessary to solve practical problems
without the destruction of the integrity of trees (Yang et al.
2005b). The earliest method was to take an increment core from
the trees, which was considered nondestructive for its life and to
develop different nondestructive techniques for studying main
physical characteristics of trees.
Researchers have used a number of techniques to investigate
the detection of defects in wood. Conners et al. (1983) used a
thresholding technique to locate surface defects on gray-scale
images of hardwood lumber. Butler et al. (1989) reported that the
use of RGB color information increased pixel-based defect de-
tection accuracy for some subtle defects by over 20 percentage
points compared to gray-scale accuracy. Sobey and Semple
(1989) used the Conners et al. (1983, 1984) gray-scale approach
on radiata pine (Pinus radiata) lumber, but the measures were
calculated for larger local areas of fixed size and only included
mean, variance, and kurtosis. They reported an overall defect
detection accuracy of 95% but clear wood accuracies of only
75%−80%. Sobey et al. (1989) then followed up that study with
a neural network-based classifier, but only overall classification
rates were given.
With the development of wood nondestructive testing tech-
Foundation project: This paper was supported by the project “Devel-
opment of Portable NDT Instrument (2002(39-1))” sponsored by Na-
tional Forestry Administrative Bureau of China
Received: 2007-05-06 ; Accepted: 2007-06-20
© Northeast Forestry University and Springer-Verlag 2007
Electronic supplementary material is available in the online version of
this article at http://dxdoi.org/10.1007/s11676-007-0045-5
Biography: WANG Li-hai (1960- ), Male, professor in Northeast For-
estry University, Harbin 150040, P. R. China. Correspondence author
(E-mail: lihaiwang@yahoo.com )
Responsible editor: Zhu Hong
nology, stress wave technology is widely used in fields of wood
industry due to its lower cost in comparison with other technolo-
gies including CT and x-rays, etc. In recent years, stress wave
technology has been used to test wood defect with stress wave
sensors simultaneously (Wang et al. 2002). This paper studied
the effects of sensor quantity on the fitness and error of testing
images by using stress wave.
Principle of log defect testing by using stress wave
The general principle is that a sensor is flipped to produce a
stress wave propagated in wood and received by other sensors.
The time interval in the process is recorded to calculate the
propagation speed of stress wave between two sensors. Stress
wave propagation in log is a dynamic process that is directly
related to the physical and mechanical properties of log. Gener-
ally, wave propagates more rapid in high density hard wood and
slower in low density soft wood (Yang et al. 2005a). By measur-
ing wave transmission time in the radial direction, the internal
condition of the log can be fairly accurately evaluated. This
measured time can be used as a predictor of the physical condi-
tions inside the log when it is converted to wave speed (Lin et al.
2005).
Experiment methods
Experiment hypotheses
Ten logs in four tree species with two kinds of defects collected
from Dailing Forestry Bureau, Heilongjiang Province were in-
vestigated in this study. The circumference of log disc is sup-
posed to be a whole circle. All kinds of log defects are supposed
to be normal geometric shape. Hollow is a normal circle, and
crack is a triangle or rectangle.
Experiment conditions
Testing work was carried out inside laboratory at temperature of
20 °C and humidity of 72%.
Stress wave testing apparatus, Arbotom, was imported from
Germany, and ST-85 digital apparatus of measuring log humidity,
WANG Li-hai et al. 222
and computer were employed in the testing.
The whole log with natural defects is earmarked in number 1,
number 2, until to number 10 (Table 1).
Table 1. The general characters of ten log samples
Serial
number
Assortments
Log mois-
ture content
(%)
Log disc
diameter
(cm)
Defect
category
Defect area
(cm2)
No.1 Aspen 8.5 36.8 Hollow 122.66
No.2 Aspen 8.6 33.74 Hollow 153.86
No.3 Birch 18.3 38.83 Hollow 196
No.4 Birch 17.5 30.24 Cracks 75.63
No.5 Aspen 9.3 31.96 Cracks 34
No.6 Korean Pine 15.5 31.19 Cracks 30.63
No.7 Korean Pine 13.5 21.96 Cracks 28
No.8 Korean Pine 13.3 31.83 Cracks 26.02
No.9 Ash 10.5 35.82
Hollow and
Cracks
104.25
No.10 Ash 10.5 39.79
Hollow and
Cracks
314
Experiment process
The tested log stands at one place where it is 1 or 2 meters away
from computer and Arbotom testing apparatus. The pins are put
into the circumference of log and kept on one level. Sensors are
hanged on the pins orderly keeping sensors and pins perpendicu-
lar (to the horizontal plane). The arcs between the pins are meas-
ured and recorded.
The apparatus are connected, and the work state must be nor-
mal. Detailed connecting methods are omitted.
Small hammer is used to hit the shock bolt of each sensor
about 8 to 10 times for reducing the random error. The data are
observed and the error data should be deleted. Then the sensors
with error data are hit again until error rate of all data under 3%.
Meanwhile line graph and surface graph originate from Arbotom
testing apparatus are analyzed.
The sensors are removed from the log, the pins are fixed again
and sensor quantity is changed. This experiment is repeated until
to end.
Experiment results
Surface graph
The testing log images of No.9 (Fig. 1) are taken as example to
describe the experiment process. These images are tested by
Arbotom software in different sensors, and intensified by Photo-
shop software.
Fig. 1 Cross section picture of log sample No. 9
Fig. 2 The surface graph of log sample No. 9 with 4, 6, 8, 12 sensors respectively
Journal of Forestry Research, 18(3): 221−225 (2007) 223
In Fig. 2, the dark color areas are defect areas. It is clear that
area and location of the tested defects approach actual defects
gradually with sensor quantity increasing.
Experiment data
The size of graph, i.e. the length of graph (L) and the width of
graph (W) can be shown from the graph originated from Ar-
botom. If the size of graph is known, the defected area tested
could be calculated. The calculate steps are:
(1) The single grid area is calculated by using the length of
graph (L) and the width of graph (W).
(2) The number of grids of defects is counted for N.
(3) The tested defect area is calculated by multiplying single
grid area and defect grid number N (Table 2).
Table 2. Tested defect area of ten log samples cm2
Sensor quantity No. Area
3 4 5 6 8 10 12
Sd 3.98 6.08 6.08 5.43 6.08 5.75 6.08
N 51 35 31 28 24 26 21 No.1
St 202.98 212.8 188.48 152.04 145.92 149.5 127.68
Sd 4.46 6.84 6.84 5.83 6.84 6.44 6.84
N 50 34 32 33 30 27 25 No.2
St 223 232.56 218.88 192.39 205.2 173.88 171
Sd 4.38 6.76 5.76 5.89 6.41 6.41 6.76
N 64 60 54 51 40 34 31 No.3
St 280.46 405.2 310.8 300.56 256.4 218.05 209.56
Sd 4.94 4.00 3.48 3.6 4 4 4
N 43 35 34 27 25 22 19 No.4
St 212.5 140 118.32 97.2 100 88 76
Sd 5.70 4.40 4.13 4.13 4.27 4.41 4.27
N 42 50 19 14 22 9 8 No. 5
St 239.50 220.44 78.47 57.82 93.94 39.69 34.16
Sd 4.00 5.69 5.15 5.14 5.50 5.50 5.69
N 41 26 25 26 15 9 6 No.6
St 163.76 147.85 128.70 133.54 82.54 49.53 34.12
Sd 3.99 5.98 5.19 5.16 5.98 5.70 5.98
N 33 18 15 15 11 7 5 No.7
St 131.59 107.56 77.78 77.41 65.78 39.92 29.88
Sd 4.15 6.06 5.15 5.68 5.68 5.68 6.06
N 29 17 17 12 9 8 5 No.8
St 120.46 103.01 87.51 68.17 51.12 45.44 30.29
Sd 6.9 5.76 4.84 4.96 5.76 5.44 5.76
N 50 50 42 31 27 23 19 No.9
St 345 288 203.09 153.76 155.52 125.12 109.44
Sd 4.67 7.11 6.08 6.22 6.22 6.76 7.11
N 93 78 57 58 54 50 46 No.10
St 434 554.67 346.56 360.89 336 337.78 327.11
Notes: N is the number of defect grids. Sd is single grid area and St is the tested defect area.
From Table 2, the number of defect grid drops with sensor
quantity increasing, and the tested defect area also reduces
gradually, which approaches the actual defect of samples gradu-
ally as a result. While closing to the actual defect, the tested de-
fect area becomes smaller. But there are some data that do not
conform to this rule. Because the defects of samples are very
complex, there are errors during the experiment.
Analysis of experiment results
Image fitting degree and error rate
For analyzing the effects of sensor quantity on the measurement
accuracy of log defects, stress wave is used by setting two in-
dexes: image fitting degree and error rate (Yang et al. 2002).
Image fitting degree reflects the degree of image matching by
using the ratio of actual defect area to tested defect area. zS is
actual defect area,
tS is tested defect area, T is image fitting
degree. There is
%100×=
t
z
S
ST (1)
Error rate reflects the deviation degree of actual defect area to
tested defect area. V is error rate. The math formula is
z
zt
S
SS
V
−= (2)
From above two formulas, accuracy of stress wave testing is
higher when image fitting degree approaches 100% and error
rate approaches 0.
Effects of sensor quantity on image fitting degree and error rate
Image fitting degree and error rate of ten samples are calculated
according to the defects area and the equation (Table 3).
The image fitting degree approaches 100% and error rate ap-
WANG Li-hai et al. 224
proaches 0 gradually with sensor quantity increasing (Table 3).
The accuracy of stress wave testing for log defect also increases
gradually with sensor quantity increasing.
Table 3. The image fitting degree and the error rate of ten samples in different sensor quantity
Area Sensor quantity
No.
3 4 5 6 8 10 12
St 202.98 212.8 188.48 152.04 145.92 149.5 127.68
Sz 122.66 122.66 122.66 122.66 122.66 122.66 122.66
T1 60% 57% 65% 81% 84% 82% 96%
No.1
V1 0.65 0.73 0.53 0.24 0.19 0.22 0.04
St 223 232.56 218.88 192.39 205.2 173.88 171
Sz 153.86 153.86 153.86 153.86 153.86 153.86 153.86
T2 69% 66% 70% 79% 75% 88% 90%
No.2
V2 0.45 0.51 0.42 0.25 0.33 0.13 0.11
St 280.46 405.2 310.8 300.56 256.4 218.05 209.56
Sz 196 196 196 196 196 196 196
T3 70% 48% 63% 65% 76% 90% 94%
No.3
V3 0.43 1.07 0.59 0.53 0.31 0.11 0.07
St 212.5 140 118.32 97.2 100 88 76
Sz 75.625 75.625 75.625 75.625 75.625 75.625 75.625
T4 36% 54% 64% 77% 76% 86% 99%
No.4
V4 1.81 0.85 0.56 0.29 0.32 0.16 0.005
St 239.50 220.44 78.47 57.82 93.94 39.69 34.16
Sz 34 34 34 34 34 34 34
T5 14% 15% 43% 59% 36% 86% 99%
No.5
V5 6.04 5.48 1.30 0.70 1.76 0.16 0.005
St 163.76 147.85 128.70 133.54 82.54 49.53 34.12
Sz 30.63 30.63 30.63 30.63 30.63 30.63 30.63
T6 19% 21% 24% 23% 37% 62% 90%
No.6
V6 4.35 3.83 3.20 3.36 1.69 0.62 0.11
St 131.59 107.56 77.78 77.41 65.78 39.92 29.88
Sz 28 28 28 28 28 28 28
T7 21% 26% 36% 36% 43% 70% 94%
No.7
V7 3.69 2.84 1.78 1.76 1.35 0.425 0.07
St 120.46 103.01 87.51 68.17 51.12 45.44 30.29
Sz 26.02 26.02 26.02 26.02 26.02 26.02 26.02
T8 22% 25% 30% 38% 51% 57% 86%
No.8
V8 3.63 2.95 2.36 1.62 0.96 0.75 0.16
St 345 288 203.09 153.76 155.52 125.12 109.44
Sz 104.25 104.25 104.25 104.25 104.25 104.25 104.25
T9 30% 36% 51% 68% 67% 83% 95%
No.9
V9 2.31 1.76 0.94 0.47 0.49 0.20 0.05
St 434 554.67 346.56 360.89 336 337.78 327.11
Sz 314 314 314 314 314 314 314
T10 72% 57% 91% 87% 93% 93% 96%
No.10
V10 0.38 0.77 0.10 0.15 0.07 0.07 0.04
Regression analysis
Calculation on average image fitting degree and average error
rate
For reducing error, the average values of image fitting degree
and error rate of ten samples are used as final value, and then the
relationships between them are analyzed by SPSS software.
From Table 4 and Fig.3, for logs with diameter ranging from
20 to 40 cm, at least 12 sensors were needed to meet the re-
quirement which ensure a high testing accuracy of roughly 90%
of fitness with 0.1 of error rate. And 10 sensors are recom-
mended to judge the possible locations of defects and 6 sensors
are sufficient to decide whether there are defects or not.
Table 4. Average image fitting degree and average error rate with
different sensor quantity
Sensor quantity Item
3 4 5 6 8 10 12
T 41.3% 40.5% 53.7% 61.3% 63.8% 79.7% 93.9%
V 2.374 2.079 1.178 0.937 0.747 0.285 0.066
Regression analysis by using SPSS software
SPSS softwares were used to conduct the regression analysis
between sensor quantity and image fitting degree or error rate
Journal of Forestry Research, 18(3): 221−225 (2007) 225
and the analysis results are shown as following:
The correlation coefficient (R) between image fitting degree
and sensor quantity is 0.983, and the regression equation is
0 .2 2 0 .0 5 8y x= + , { }3 , 4 , , 1 2x = " (3)
where, y is the image fitting degree and x is the sensor quantity.
The correlation coefficient (R) between error rate and sensor
quantity is 0.939, and the regression equation is
2 .7 8 7 0 .2 4 7y x= − , { }3 , 4 , , 1 2x = " (4)
where, y is the error rate and x is the sensor quantity. The image
fitting degree and error rate between actual defect and tested
defect with different sensor quantity are estimated by the regres-
sion equation.
Fig. 3 The relation between sensor quantity and average image fitting degree or average error rate
Conclusions
The correlation between sensor quantity and image fitting degree
is positive and notable. The regression equation between them is
0 .2 2 0 .0 5 8y x= + , { }3 , 4 , , 1 2x = " .
The correlation between sensor quantity and error rate is nega-
tive and notable. The regression equation between them is
2 . 7 8 7 0 . 2 4 7y x= − , { }3 , 4 , ,1 2x = " .
For logs with diameter ranging from 20 to 40cm, at least 12
sensors were needed to meet the requirement which ensure a
high testing accuracy of roughly 90% of fitness with 0.1 of error
rate. And 10 sensors are recommended to judge the possible
locations of defects and 6 sensors are sufficient to decide
whether there are defects or not.
Sensor quantity has great effects on the precision of log de-
fects testing by using stress wave. Appropriately increasing sen-
sor quantity can enhance image fitting degree and reduce error
rate, and finally enhance the precision of stress wave testing.
However, it is not enough to enhance the precision of stress wave
testing by increasing sensor quantity. Because it is influenced by
various factors, for example power of hitting the shock bolt using
small hammer, position of sensor on the level, various assort-
ments, percentage of moisture and density etc. Therefore further
research work should be beneficial on testing the log inner de-
fects by using stress wave.
References
Butler, D.A., Brunner, C.C., Funck, J.W. 1989. A dual-threshold image
sweep-and-mark algorithm for defect detection in veneer. Forest Products
Journal, 39 (5): 25−28.
Conners, R.W., McMillin, C.W., Lin, K., et al. 1983. Identifying and locating
surface defects in wood: part of an automated lumber processing system.
Pattern Analysis and Machine Intelligence, 5 (6): 573−583.
Lin Wenshu, Yang Huimin, Wan Lihai. 2005. Comparative study on ultrasonic
and stress wave for nondestructive test of wood defects. Forestry Science
and Technology, 30(2): 39−41.
Sobey, P.J., Semple, E.C. 1989. Detection and sizing visual features in wood
using tonal measures and a classification algorithm. Pattern Recognition,
22 (4): 367-380.
Wang Lihai, Yang Xuechun. 2002. Current situation of research on the
non-destructive testing technique for wood defects. Forestry Science and
Technology, 27(3): 35−38.
Yang Xuechun, Wang Lihai. 2005a. Impacts of structure defaults on propaga-
tion parameters of stress wave in wood of korean Pine. Journal of North-
east Forestry University, 33(1): 30−31.
Yang Xuechun, Wang Lihai. 2005b. Study on the Propagation Theories of
Stress Wave in Log. Scientia Silvae Sinicae, 41(5): 132−138.
Yang Xuechun, Wang Lihai. 2002. Research progress of testing timber char-
acteristics by using stress wave technique. Forest Engineering, 18(6):
11−12.
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
90.00%
100.00%
3 4 5 6 8 10 12
Sensor quantity
Im
ag
e
fit
tin
g
de
gr
ee
0
0.5
1
1.5
2
2.5
3 4 5 6 8 10 12
Sensor quantity
Er
ro
r r
at
e
Chinese Abstracts 3
黑曲霉对二氯甲烷提取物进行抑菌性能测试,结果表明该提
取物对黑曲霉没有抑菌能力。图4表2参29。
关键词:日本花柏;心材外缘;二氯甲烷提取物;气质联机;
抑菌.
CLC number: Q946.8 Document code: A
Article ID: 1007−662X(2007)03-0042-08
07-03-009
ACQ 防腐剂对扭叶松蓝变部分木材力学性能的影响/江京辉
任海青 吕建雄 骆秀琴 吴玉章(中国林业科学研究院木材工
业研究所,国家林业局木材科学与技术重点实验室,北京
100091)//Journal of Forestry Research.-2007, 18(3): 213−216.
本文利用三种不同浓度 ACQ 防腐剂对扭叶松蓝变木材
进行浸注处理,其浓度分别为 1.2%、2.0%和 2.8%。研究其
抗弯弹性模量、抗弯强度、冲击韧性和顺纹剪切强度(弦面)
与未处理蓝变木材相应力学性能的差异,测试
标准
excel标准偏差excel标准偏差函数exl标准差函数国标检验抽样标准表免费下载红头文件格式标准下载
参照
GB1927~1943-91。研究结果显示,经浸注处理后的试样均达
到美国 AWPA 标准 UC4A 等级规定的药剂保持量;ACQ 防
腐处理大约降低了 20%扭叶松蓝变木材的冲击韧性,与未防
腐处理试样对比,在 0.01水平上有显著差异,但不同浓度间
差异不显著;三种浓度 ACQ处理间以及与未处理的扭叶松蓝
变木材的抗弯强度、抗弯弹性模量和顺纹剪切强度差异不显
著;随着 ACQ浓度的降低,冲击韧性、抗弯强度、抗弯弹性
模量和顺纹弦面剪切强度有所增大,但影响都很小。图 4 表
10 参 6。
关键词:扭叶松,蓝变处理材,非处理材,冲击韧性,抗弯
强度,抗弯弹性模量,顺纹剪切强度(弦面)
CLC number: S781.2 Document code: A
Article ID: 1007−662X(2007)03-0213-04
07-03-010
不同水分条件下紫藤叶片光合作用的光响应/张淑勇(中国林
业科学研究院林业研究所, 北京 100091;国家林业局林木
培育重点实验室,北京 100091),夏江宝(滨州学院黄河三角
洲生态环境研究中心,滨州 256603),周泽福(中国林业科学
研究院林业研究所, 北京 100091;国家林业局林木培育重
点实验室,北京 100091),张光灿(山东农业大学林学院,泰
安 271018) //Journal of Forestry Research.-2007, 18(3):
217−220.
通过测定 2 年生紫藤叶片气体交换参数的光响应,确定
紫藤正常生长发育所需的土壤水分条件。结果表明:紫藤的
光合速率(Pn)、蒸腾速率(Tr)及水分利用效率(WUE)对
土壤湿度和光照强度的变化具有明显的阈值。维持紫藤同时
具有较高 Pn和WUE的土壤湿度范围,在体积含水量(VWC)
为 15.3%~26.5%,其中最佳土壤湿度为 23.3%。适宜的土壤
水分条件下,紫藤光饱和点在 800µmol·m-2·s-1 以上,在水分
不足(VWC为 11.9%,8.2%)或渍水(VWC为 26.5%)时,
光饱和点低于 400µmol·m-2·s-1。此外,光响应曲线表明,随着
光合有效辐射(PAR)增加到一特殊点,气孔限制值(Ls)和
胞间 CO2浓度出现相反的变化趋势。这个点的光合有效辐射
称为光合作用由气孔限制转变为非气孔限制的转折点。并且
不同水分条件下的转折点各不相同,当 VWC 为 28.4%,
15.3% 11.9%和 8.2%,转折点分别为 600, 1000,1000 and 400
µmol.m-2.s-1。总之,紫藤通过对自身生理机能的调节,对水
分胁迫具有较高的适应能力。图 6参 26。
关键词:净光合速率;土壤湿度;光合有效辐射;水分利用
效率;紫藤
CLC number: Q945.11 Document code: A
Article ID: 1007−662X(2007)03-0217-04
07-03-011
传感器数量对应力波检测原木内部缺陷精度的影响/王立海,
徐华东, 周次林, 李莉, 杨学春 (东北林业大学,哈尔滨
150040) //Journal of Foresetry Research.-2007, 18(3): 221−225.
木材无损检测技术