Agriculture Reference
In-Depth Information
7.6.4 M
ACHINE
V
ISION
U
SED
IN
Y
IELD
E
STIMATION
Annamalai and Lee (2003) developed an image processing algorithm to locate citrus
fruits in a canopy image. By studying the pixel distribution of hue and saturation
color plane, the thresholds (Figure 7.43) to separate different classes (citrus fruits,
leaves, and background) were estimated. The algorithm consisted of the following:
(1) color image was transformed to binarized image through the threshold in HSI
color image; (2) then erosion and dilation were applied to remove noise; (3) after that,
the gap within a fruit was removed by applying dilation and erosion; and, finally, (4)
the number of fruits was counted using blob analysis. The result showed that the cor-
relation coefficient was 0.76 for the regression analysis between the manual method
and machine vision algorithm. Annamalai et al. (2004) made an improvement on the
previous algorithm by adding a luminance component to the threshold to make it less
dependent on the brightness level.
Swanson et al. (2010) provided methods to calculate tree canopy volume, density,
immature orange counts, and mature orange counts. They used intensity profiles to
detect green oranges. For mature oranges, they first converted the image into LAB
color space. Then, the oranges were separated from the background by calculating
the minimum distance from each pixel to the sample region means. Finally, mor-
phological operators and watershed transform were used to filter and segment fruits
cluster (Figure 7.44).
Han and Burks (2010) used a Plucker coordinates system to reconstruct a 3-D
image scene of a citrus canopy using monocular vision. A single video camera was
mounted on a robot manipulator to capture multisequential views of citrus canopy.
The centers of tree leaves were detected as the leaf features. These feature points
were used for feature matching in the SURF method. After feature matching, two
300
Threshold
Citrus
Leaf
Background
250
200
Threshold
150
100
50
0
0
50
100
150
Hue (gray level)
200
250
300
FIGURE 7.43
Pixel distribution of three different classes from 31 calibration images in the
HS color plane. (From Annamalai, P., Lee, W.S., Citrus yield mapping system using machine
vision,
ASAE Annual Intl. Meeting
, 2003. With permission.)
