You are on page 1of 4

2008 International Conference on Computer Science and Software Engineering

Grading Method of Leaf Spot Disease Based on Image Processing

Shen Weizheng1 Wu Yachun1 Chen zhanliang2 Wei Hongda3


1Department of computing, Engineering College, Northeast Agricultural University,
Harbin150030, China E-MAIL: alienmars@163.com
2Department of agricultural electrization and automatization, Engineering College,
Northeast Agricultural University, Harbin150030, China
3Anmin countryside, Zhaodong151103, China

Abstract research, such as agricultural products quality


inspection and classification, the crop growth state
Since current grading of plant diseases is mainly monitoring, plant disease and insect pests
based on eyeballing, a new method is developed identification, and other agricultural robot[8,9]. With
based on computer image processing. All influencing the rapid development of computer hardware and
factors existed in the process of image segmentation software technology, the application of image
was analyzed and leaf region was segmented by using processing technology in the agricultural research has
Otsu method. In the HSI color system, H component made significant development.
was chosen to segment disease spot to reduce the
disturbance of illumination changes and the vein. 1 Classification Principle
Then disease spot regions were segmented by using
Sobel operator to examine disease spot edges. Finally The severity extent of the plant leaves diseases is
plant diseases are graded by calculating the quotient commonly measured by the ratio of lesion area and
of disease spot and leaf areas. Researches indicate leaf area ratio. Adopting image processing method to
that this method to grade plant leaf spot diseases is measure can be expressed as the following formula
fast and accurate.
p 1 1
Key words: Image segmentation; Disease
grading; Computer application
Ad ( x , y ) R d ( x , y ) R d
t= = =
0 Introduction Al p 1 1
( x , y ) Rl ( x , y ) Rl
Plant diseases can cause significant reduction in
rops and lead to poor quality of agricultural
Ad Lesion Region Area;
products[1], thus influence economy and people's life. Al Leaf Region Area;
The mensuration of the extent plant disease is the
essential item in the research of disease control, yield
p Unit Pixel Expressed Area;
loss estimate, disease resistance and breeding, as well Rd Lesion Region;
as disease catch mechanism[1-7]. Although there is an
industrial recognized corresponding standard to grade
Rl Leaf Region.
the leaf spot disease[4-7], the naked eye observation Unit pixel in the same digital image represente the
method is mainly adopted in the production practice. same size, so ratio t can be obtained by segmenting
Because of the difference of personal knowledge and lesions and leaves and calculating pixel
practical experience, the same samples are classified
into different grades by different experts. Therefore,
number 1 of lesion region and ( ) 1 of leaves
( x, y )Rd x, y Rl
the result is usually subjective and it is impossible to
region in the binary image. Then according to disease
measure the disease extent precisely. Although grid
classification standard consult table the final severity
paper method can be used to improve the accuracy, it
level can be achieved.
is seldom used in practice due to cumbersome
operation process and time-consuming. Therefore
looking for a fast and accurate method to measure 2 Image Acquisition
plant disease severity is of great realistic significance.
Since the late 1970s, computer image processing This study takes gray leaf spot disease of soybean
technology is applied in the agricultural engineering as example for illustrating the plant disease extent
grading method. Infected leaves are placed flat on a

978-0-7695-3336-0/08 $25.00 2008 IEEE 491


DOI 10.1109/CSSE.2008.1649
white background. The optical axis of digital camera correspond to the leaf region with lower gray value.
is perpendicular to the leaf plane to shoot images, Due to the uneven illumination when collecting
which are deposited in the computer for future use. pictures in the field, there are a few splits between the
bachground region and its corresponding wave crest.
3 Image Segmentation While the gray value between leaves and background
vary greatly (distance between wave crest>100),
Image segmentation is not only the basic threshold segmentation is not impacted. Otsu is used
technology of image processing and prophase vision, to segment images[13,14], which is an overall
but also the important component of most image automatic threshold selection mothod imposed by the
analysis and visual system[10]. Image segmentation is Japanese scholarDa Jin. Otsu is derived from the
to separate the different regions with special least binary multiplication principle, which basic idea
significance in the image. These regions do not is if an image is formed by an object and background
intersect each other, and each region should meet with different gray values, some gray value is
consistency conditions in specific regions[11]. In this adopted as threshold to segment the image into two
study two-step division is implemented to obtain the groups--the ogject and background and calculate the
leaves and lesion areas. variance between the two groups. When the largest
variance is achieved, this gray value is used as
3.1 Leaf regional segmentation threshold for image segmentation. After image
segmentation, the binary image contained leaf
In order to reduce computation by obtaining region Rl is obtained by region filling and eliminating
region and leaf image and narrowing the regions of holes in the white region. After scanning the image
interest , the background should be removed firstly. from top to bottom and from left to right, the number
The formula of graying the collected RGB color
image [12] is:
of pixel in the white non-zero region 1
( x, y )Rl
is
I = 0.3R + 0.59G + 0.11B 247784.
Image gray statistical histogram basically shows
bimodal characteristics. The left wave crest

Gray Image Gray Histogram Binary Image of Leaf Region


Fig.1 Segmentation of leaf region

In different stages of the disease, under the


influence of light, water, nutrition, and so on, the
3.2 Lesion Segmentation lesion manifest various symptoms, which brings
difficulties to the segmentation.
The accurate segmentation of lesion region Based on the above factors, the image is
determines the success or failure of the experiment. transformed from RGB color space to HSI color
The following aspects influence the experimental space[15], which is more suitable for visual
process: characteristics of human beings. Since the brightness
1) Illumination Change component is independent of the color component
In light of in this study the image can be collected and the vision of human beings is more sensitive to
isolated or alive and it is very difficult to control light hue compared to saturation, the color component can
in the process of wild living collection, the natural be good to eliminate glare, shadows and other light
light is usually adopted. Therefore imaging quality is factors during color image segmentation. And the
influenced by light changes. similar gray value of the shallow color vein and the
2) Influence of Vein leaf in color component can decrease the interference
The vein color is usually shallower than the leaf of the vein in the follow-up lesion image
color and in the early stages of illness the green color segmentation to a large extent. Although the lesion
often fade. Therefore in the course of lesion characteristics are varied, the boundaries between the
segmentation, the phenomenon of the wrong lesion and the healthy part are obvious and the color
segmentation often exist. of the lesion change frequently. Hence its
3) Instable Lesion Characteristics corresponding gray gradient changes will be so
dramatic that it can be detected by edge detection

492
operator. Meanwhile the larger leaf margin gray
dispersion at the junction of the leaf blade and 1 0 1 1 2 1
background can be detected by gradient operator. The M 1 = 2 0 2 M 2 = 0
0 0
leaf area is eroded to narrow the interest area by
using the circular structure element within the radius 1 0 1 1 2 1
of three. To multiply the binary image after corrosion Output Gray Value
by H component can achieve the target image with
[( )]
1
the black background. Sobel operator and the use of
H component convolution to obtain gradient image. ) (
2
G = M 1T F + M 2T F
2 2

Sobel operator with the form of vector can be


decomposed into the following two models [16].

H Component Wiped Off Background Sobel Gradient Image Threshold Segmented Image

Region Filled Image Lesion Region Binary Image Detected Lesion Image
Fig.2 Segmentation of Disease Region

A suitable threshold binary gradient image is


chosen. From the figure it can be seen that the lesion T2 0.01 < t 0.05
edge can be well detected. Because the edge is
closed, the region can be filled to obtain integrated
lesion image. It can be seen from the figure that
T3 0.05 < t 0.2
although some veins can be detected, they can be
dealt as noise since they are scattered and their unit T4 0 .2 < t 0 .5
area is small compared to the lesion case. To select
the appropriate circular structure elements to unseal
to obtain the final binary image of lesion. After T5 0 .5 < t 1
scanning the number of pixels 1 in the lesion
( x , y )Rd
region is 6602. Thus it can be calculated that the ratio 4 Conclusion
t of the lesion and leaf area is 0.027. According to
the grade table(as the following table 1) of the Plant diseases happen under natural conditions and
soybean gray leaf spot disease provided by the their symptoms vary significantly under different and
literature [4], the infected level of the leaf is T2 in the different stages of disease. Whether the image
after programming to check up the table. is segmented correctly or not determines the success
or failure of the study. With the use of Otsu leaves
Tab.1 Rank Table of Soybean Leaf Spot Disease are segmented. The white background is more
suitable for image segmentation proved by
Ratio of Lesion to Leaf experiments. Because of the variability of lesion
Disease Level
Area symptoms, threshold segmentation is not suitable for
use. This paper presents to segment lesion accurately
T0 t =0 by using Sobel operator to extract lesion edge and
operations as region fill, morphology open operation
in the H component.
T1 0 < t 0.01 The usage of image processing technology for
plant disease degree grading eliminates the

493
subjectivity of traditional classification methods and and treatment of soybean [J]. Soybean Science. 1989,
human-induced errors. Thus the estimation credibility 8(1):75-84.
is improved and accurate data are provided for [6] Liu Xuemin, Li Changyou, Zhang Minghou.
disease studies. The method is also convenient, which The grading standards of Soybean gray leaf spot
simply needs computers, digital cameras with the diseased leaves spots of the severity[J]. Soybean
combination of necessary software programs to Science. 1991, 10(4):330-334
realize for the disease batch grading. [7] Cao YuePing, Yang Qingkai. Research on the
standard anti-flu of Soybean gray leaf spot disease.
5 Acknowledgements Soybean Science. 2002, 21(2):113-116.
[8] He Dongjian, Zhang Hailiang, Ning Jifeng.
We acknowledge support of this work by National Automated computer vision technology applications
Science and Technology Support Programme in agriculture the field [J]. Transactions of the
(2006BAJ09B0702). We acknowledge support of this Chinese Society of Agricultural Engineering. 2002,
work by Computer Department of Engineering 18(2)171-175.
College in Northeast Agricultural University. [9] Fang Ruming, Cai Jianrong, Xu Li. Computer
image processing technology and its application in
agricultural engineering[M]. Beijing: Tsinghua
6 References University Press, 1999.
[10] Sahoo P K, Soltani S, Wong A K C, et al. A
[1] Li Haiyan, Liu Xiruo, Zhen Hongjie. Research survey of thresholding techniques [J]. Computer
on quality and product losses of soybean by grey leaf Vision, Graphics, and Image Processing. 1988, 41(2):
spot[J]. Chinese Journal of Oil Crop Sciences. 2005, 233-260.
27(3):66-69. [11] Lin Kaiyan, Wu Junhui, Xu Lihong. Color
[2] Shi Zhenzi, Guo Yuren, Lu Baoli. Rice false image segmentation method Summary[J]. Journal of
smut classification standards and preliminary Image and Graphics. 2005, 10(1):1-10.
determination of yield loss [J]. Journal of Shanghai [12] Zhou Changfa. Visual C++ image processing
Jiaotong University. 2003, 21(2):152-155. programme[M]. Beijing: Electronics Industry
[3] Huang Chunyan, Ma Shumei, Zhang Zengmin, Publishing House, 2004.184-187.
Zhu Chuanying, Li Baoying. Prevalence rate of the [13] Otsu N. A threshold selection method from
diseased leaves and the severity of relations discussed gray-level histogram [J].IEEE Trans Actions on SMC,
about Soybean gray leaf spot disease. Acta 1979(9): 652-655.
Phytopathologica Sinica. 1999, 29(1):73-76. [14] Liu Jianzhuang, Li Wenqing. Two-
[4] Wu Xiuhong, Wen Jingzhi, Ma Shumei. dimensional Otsu automatic threshold segmentation
Soybeans of new anti-gray leaf spot disease situation method of the gray image[J]. Acta Automatica Sinica,
analysis in Heilongjiang Province. Soybean Science. 1993, 19(1): 101-105.
2004, 23(2):114-117. [15] Zhang Shuji. Image processing and
[5] Yao Haoran, Li Yong, Li Jing. Fungicide new analysis[M]. Beijing: Tsinghua University Press,
formulations - "40 percent of carbendazim hanging 1999, 19-21.
plastic" research on gray leaf spot disease prevention [16] Rafael C. Gonzalez, Richard E. Woods.
Digital image processing[M]. Beijing: Electronics
Industry Publishing House. 2003.

494

You might also like