Development of 360-degree view imaging using L*a*b* color space for fresh fruit bunch identification
The palm oil industry is well known as a significant agricultural industry in terms of economic benefit for several tropical countries, particularly in Malaysia (Yoshizaki et al., 2013). Total amount of bunches in each of the tree is an important aspect in oil palm harvesting process. In every cy...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2020
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/98047/1/FK%202021%2028%20UPMIR.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The palm oil industry is well known as a significant agricultural industry in
terms of economic benefit for several tropical countries, particularly in
Malaysia (Yoshizaki et al., 2013). Total amount of bunches in each of the
tree is an important aspect in oil palm harvesting process. In every cycle
of harvesting operation, farmer does not have any information on how
many bunches and which oil palm tree will be harvested. By introducing
360o view imaging model for bunch identification, number of Fresh Fruit
Bunch (FFB) can be identified in a certain plantation area. Black bunch
census was done manually to estimate yield. This can be improved by
video acquisition using a high-resolution 360o camera integrated with an
image processing software for video image processing to get a 360o view
of each tree. Based from the standard planting pattern, it is a time
consuming process to circle each tree to acquire the 360o view of each
tree. In order to overcome this, a new method of data collection was
established with the execution of All-Terrain Vehicle (ATV) between rows
in the plantation area for video acquisition. The video recorded was
processed in a software in order to construct a 360o view of each oil palm
tree for further FFB identification process could be done. Image extraction
was done using the processing software by referring the data from range
sensor installed at the ATV throughout the data collection process in the
oil palm plantation. After useful image were extracted, MATLAB software
was programmed to process all the selected images for the detection of
FFB of each tree. Image pre-processing was conducted where errors in
the image were corrected to detect the oil palm bunches. In order to
present an appropriate format of image processing system, the RGB
images were converted into grayscale images. Image segmentation was
done based on a threshold value of L*a*b* to separate between canopy
and trunk, fruit bunches and background image. The features were
extracted from each pixel of the RGB image. As a result, a new method for video acquisition is established as well as a processing method for
bunch counting for large scale plantation area.
In this research, mean value for L*a*b* color space was determined by
using 90 images samples for image threshold in order to identify the FFB
on tree crown. Using L*a*b* color space, image was threshold to identify
black and both red and black FFB. In this research, image verification was
done by using the mean L*a*b* value for black bunch identification. Model
threshold verification for 48 samples of images resulted with Coefficient of
Determination, R2 of 0.8029 to identify black bunch on each tree crown.
The outcome for this research will help to fully automate the process for
bunch identification in the future. |
|---|