Image processing based weed detection in lawn Ukrit Watchareeruetai*,Yoshinori Takeuchi,Tetsuya Matsumoto,Hiroaki Kudo,Noboru Ohnishi Department of Media Science, Graduate School of Information Science, Nagoya University 1.Introduction In order to control weed in the field, a large quantity of the herbicide have been used. Using herbicide not only increases the cost of production but also makes the environment harms. To
thresholding method. In this experiment, we set the condition that the areas of weeds must have the mean and the variance less than the threshold values,
M max
(15) and
Vmax
(800)
respectively. An example of detected result is shown in Figure 3.
reduce the usage of the herbicides, hand labor may be the best way for removing weeds but it is costly and may spend a lot of time. Therefore, automated weed control system becomes an alternative solution for this problem. Based on computer vision and image processing theory, weeds in captured images will be detected and then system will spray the herbicide only over the location of detected weeds, instead of uniformly spraying on the
Fig.2.Edge image
Fig.1. Input weed image
overall area. There are some proposed systems which were tested in the cabbage fields, carrot fields[1], or tomato fields[2]. Typically, these systems start by discriminating the area of plant by using color information because the color of plant is clearly different from that of background. Then, features of plant are extracted and used to identify the known crop species in the field. Finally, Fig.3. Detected weed
the other plants are assumed to be weeds. Different from those works, we try to develop a method for
3.Summary and Future Works
detecting weeds in the lawn. For this situation, detection method
In this paper, we proposed the method using the simple image
is quite different from those works. Because of the likeliness of
processing techniques for detecting weeds in the lawn. This
the color of weed and grass, color information may not be able to
method is based on the assumption of the difference between the
be used for segmentation. Moreover, because known plant is
statistical values of weeds area and grass area. The area of weeds
grass, to identify all of grass in the lawn may be a difficult task.
can be detected by using the thresholding method. Anyway, in
In this work, we employ the simple image processing techniques,
this experiment, the threshold values were manually selected. In
such as the edge detection operator and the thresholding
practical, the proper threshold values may be changed by some
segmentation, for detecting weeds in the lawn.
factors such as the variation of light, so we need to develop a
2.Processing Method
method for choosing the suitable threshold values. Moreover,
To detect the location of weeds in the lawn, the proposed method is based on the assumption that “the area of grass should contain a lot of edges and the variance of this area should be high but the area of weeds should be quite smooth, therefore the variance of weeds area should be low”. An example of weeds
because we calculated the statistical values of each pixel in the image, in the other word we do pixel-level segmentation, it takes quite the computation time. So, we plan to solve this problem by doing coarsely segmentation, i.e. block-level segmentation, before pixel-level segmentation.
image is shown in Figure 1. Based on the above assumption, first we convert the image into 8-bit gray-scale image. Second, we
References
applied the Sobel operators to the gray-scale image and then we
(1) J. Hemming, and T. Rath, “Computer-Vision-based Weed Identification under Field Conditions using Controlled Lighting”,
get the edge image as shown in Figure 2. Then, the mean and the
Journal of Agricultural Engineering Research, 2001, Vol. 78, No. 3,
variance of each pixel will be computed from pixel values in the
233-243.
window N×N (in this work, N is 17) whose center locates at the
(2) W.S. Lee, D.C. Slaughter, and D.K. Giles, “Robotic Weed Control
pixel. Finally, we segment the area of weeds by mean of the
System for Tomatoes”, Precision Agriculture, 1999, 1(1): 95-113.