Most Suitable SNIC Parameters and Classifier Algorithm for Object-Based Classification of Rice Crop Area using Remote Sensing Images in Google Earth Engine
Geographical analysis · 14.12.2024 · 13 (2) · 71-78
DOI: 10.53989/bu.ga.v13i2.57 ↗Kurz gefasst
Fachartikel in der Zeitschrift Geographical analysis, online erschienen am 14.12.2024. Verfasst von W Khamnoi und 3 weiteren Autor:innen.
Abstract (Originalsprache)
This study aims at identifying rice crop area in Chae Chang and Buak Kang subdistricts, of San Kamphaeng District, Chiang Mai Province, Thailand through Object Based Image Analysis approach. It investigates the best fit of three elements: (1) most suitable band combination of Sentinel-1 and Sentinel-2 satellite images; (2) optimum parameters for image segmentation; and (3) best performing classifier algorithm in Google Earth Engine [GEE]. The six bands of Sentinel-1 [VV & VH from ascending and descending orbits] and Sentinel-2 [B2, B3, B4 & B8] were used in different combinations. Simple Non-Iterative Clustering [SNIC] method was applied for image segmentation. Three algorithms, Support Vector Machine [SVM], Gradient Boosting Trees [GBT] and Random Forest [RF] were tested for classification and validation. The study analyzed the outcomes of five different band combinations, thirty sets of SNIC parameters - including Compactness [Co], Connectivity [Cn], Neighborhood Size [Ns], Segment Size [Ss] - and three classifier algorithms in GEE. The highest overall accuracy of 97% and a Kappa coefficient of 0.94 was achieved by using all the six bands of the images of the two satellites [Ascending orbit of Sentinel-1] with the SNIC parameter set including Co=0.1, Cn=8, Ns=10, and Ss=5 in RF classifier algorithm. The study reveals that higher Co levels lead to more circle like segments thus unsuitable for rectangular agricultural fields. The results validated against Regions of Interest [ROI] indicate that the optimized SNIC parameters effectively delineated the big and small rice fields covering 34.64 km² [73%] of the total 47.23 km² area of the two sub-districts. The study offers a pathway for improving classification accuracy in various contexts, particularly in rice area classification. Keywords Object-Based Image Analysis, Simple Non-Iterative Clustering, Sentinel-1 and Sentinel-2 images, Support Vector Machine Classifier, Gradient Boosting Trees Classifier, Random Forest Classifier
Zitieren
APA
Khamnoi, W., Homhuan, S., Suwanprasit, C., & Shahnawaz, S. (2024). Most Suitable SNIC Parameters and Classifier Algorithm for Object-Based Classification of Rice Crop Area using Remote Sensing Images in Google Earth Engine. Geographical analysis, 13(2), 71-78. https://doi.org/10.53989/bu.ga.v13i2.57
Vancouver
Khamnoi W, Homhuan S, Suwanprasit C, Shahnawaz S. Most Suitable SNIC Parameters and Classifier Algorithm for Object-Based Classification of Rice Crop Area using Remote Sensing Images in Google Earth Engine. Geographical analysis. 2024;13(2):71-78. doi:10.53989/bu.ga.v13i2.57
BibTeX
@article{Khamnoi2024,
author = {Khamnoi, W and Homhuan, S and Suwanprasit, C and Shahnawaz, Shahnawaz},
title = {Most Suitable SNIC Parameters and Classifier Algorithm for Object-Based Classification of Rice Crop Area using Remote Sensing Images in Google Earth Engine},
journal = {Geographical analysis},
year = {2024},
volume = {13},
number = {2},
pages = {71--78},
doi = {10.53989/bu.ga.v13i2.57}
}
EUVIMED