IDENTIFICATION AND AREA ESTIMATION OF RICE FIELD FROM SATELLITE IMAGERY USING PCA AND CNN METHODS
The problem of food security is ongoing. One from many indicator that can measure the problem of food security is the availability of rice, so one way to understand the status of food security in a country is to map rice land. At present, mapping is carried out by identifying land that utilizes the...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/42912 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | The problem of food security is ongoing. One from many indicator that can measure the problem of food security is the availability of rice, so one way to understand the status of food security in a country is to map rice land. At present, mapping is carried out by identifying land that utilizes the vegetation index, where the vegetation index is derived from a decrease in satellite imagery data. The number of vegetation indices used causes the problem called the curse of dimensionality, where the problem can be overcome by dimension reduction techniques. One dimension reduction technique that is widely used is Principal Component Analysis (PCA). PCA technique will be combined with 1-Dimensional Convolutional Neural Network (1-D CNN) and some baseline classifier, namely Random Forest (RF) and Support Vector Machine (SVM) to see the effect on several performance factors, namely F1-Score value, data training time, and data prediction time. Also, the combined model was tested to measure the estimated area of paddy land and to compare it with the estimated area of paddy land that was carried out manually. From the performance factors produced, CNN combined with PCA was proven to produce the best F1-Score. Besides, the data training time is also reduced when combining PCA with all classifiers tested. Insignificant time reduction occurs for data prediction time due to a fairly small time difference. For estimation of rice land area, there is still a high enough error because the vegetation index in the training data still has similarities with the vegetation index in the rice land estimation test data, even though in the tested area there is no paddy land. Besides, field data collection needs to be done to improve the accuracy of the estimated rice area. |
|---|