Oil palm female inflorescences anthesis stages identification using selected emissivities through thermal imaging and Machine Learning
Oil palm industry seeks for a reduction of cost and environmental impact, promote sustainability and to maximize crop production and quality. In the oil palm production process, pollination is one of the main factors contributing to yield. However, oil palm pollination is facing issues with fruit...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2022
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/114868/1/114868.pdf http://psasir.upm.edu.my/id/eprint/114868/ http://ethesis.upm.edu.my/id/eprint/18193 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Oil palm industry seeks for a reduction of cost and environmental impact,
promote sustainability and to maximize crop production and quality. In the oil
palm production process, pollination is one of the main factors contributing to
yield. However, oil palm pollination is facing issues with fruit formation and filling
due to poor natural pollination. Alternatively, assisted/artificial pollination and
Wireless Sensor Network (WSN) systems have been introduced to determine
pollination readiness of the oil palms, break the reproduction cycle, and
producing new breeding material. To perform these methods, several factors are
taken into account such as the number of inflorescences to be pollinated per
hectare and if these inflorescences require the opening of bracts. These tasks
are labor-intensive and subjective to the availability of experts. All these methods
depend on manual monitoring and visual inspection with ever decreasing labor,
making farming economically inefficient and unstable. Therefore, it’s necessary
to identify the pollination stages to ensure successful assisted/artificial pollination
operation. To achieve this in digital agriculture, useful data about the different
stages of oil palms inflorescences pollination is necessary to deliver better
decision-making systems. This research studies different Machine Learning (ML)
classification and ensemble techniques for the assessment of the four pollination
stages consist of pre-anthesis I, pre-anthesis II, pre-anthesis III, and anthesis
using thermal imaging. Different ML algorithms such as Random Forest (RF), k
Nearest Neighbor (kNN), Support Vector Machine (SVM), Artificial Neural
Network (ANN) as well as an ensemble method are used on data extracted from
thermal images collected during infield oil palms pollination stages monitoring.
Thermal images are captured with a selected emissivity values of 0.96, 0.97, and
0.98 and evaluated to determine the best model performance. To apply the
above-mentioned models, there are two feature sets are utilized consisting of
endogenous features from thermal images taken with three emissivity values
and exogenous features including meteorological variables. These models’
performance is validated statistically and empirically considering the average
accuracy, recall, precision, and F-measure in classifying the present datasets.
The ensemble method on endogenous and endogenous+exogenous feature
sets from emissivity of 0.96 achieved F-measure scores of 92.68% and 93.42%
respectively and with emissivity of 0.97 resulted in 87.06% and 89.73%
respectively. However, the ensemble method on emissivity of 0.98 using
endogenous and endogenous+exogenous feature sets resulted in F-measure
score of 57.81% and 86.63%, respectively lower than that of the latter.
Ultimately, the results suggest that the proposed ML method can be utilized
effectively to accurately estimate the four pollination stages in plantations,
becoming a reliable and accurate tool in automated assisted/artificial pollination
decision making systems. The proposed detection system capable of rapid and
accurate screening and identification of oil palms inflorescences can be applied. |
|---|