Nabilah Fitriani, Dano Fadilah Amelya Rizki, Soffiana Agustin

KLASIFIKASI TUTUPAN LAHAN SAWAH DAN KELAPA SAWIT MENGGUNAKAN GLCM DAN K-NEAREST NEIGHBOR PADA CITRA UDARA

Introduction

Klasifikasi tutupan lahan sawah dan kelapa sawit menggunakan glcm dan k-nearest neighbor pada citra udara. Klasifikasi otomatis tutupan lahan sawah & kelapa sawit dari citra udara pakai GLCM & K-NN. Capai akurasi 97.30% untuk pemetaan & monitoring pertanian.

Abstract

This study aims to automatically classify rice field and oil palm land cover based on aerial imagery by utilizing the Gray Level Co-occurrence Matrix (GLCM) for texture feature extraction and the K-Nearest Neighbor (KNN) algorithm as the classification method. The dataset consists of 130 training images and 111 test images. The images were processed through cropping and grayscale conversion, followed by texture feature extraction including contrast, correlation, energy, and homogeneity. These features serve as the foundation for distinguishing the unique texture patterns of each land type. The test results show that the K parameter in KNN significantly affects the classification accuracy, with K=7 achieving the best result of 97.30%. Evaluation using a confusion matrix reinforces the effectiveness of the method in distinguishing the two land cover classes. The combination of GLCM and KNN proves to be both efficient and accurate, with great potential to be applied in automated mapping and monitoring systems, particularly in agricultural and plantation contexts.

Review

This study presents a clear and focused approach to automating the classification of rice field and oil palm land cover using aerial imagery, a topic of significant practical importance for agricultural and plantation management. The authors leverage a well-established combination of the Gray Level Co-occurrence Matrix (GLCM) for texture feature extraction and the K-Nearest Neighbor (KNN) algorithm for classification. The paper addresses a critical need for efficient and accurate land cover mapping, which can inform policy, resource allocation, and environmental monitoring, particularly in regions where these two crops dominate the landscape. The methodology is systematically described, starting with essential preprocessing steps like image cropping and grayscale conversion, which prepare the aerial imagery for texture analysis. The choice of GLCM for extracting features such as contrast, correlation, energy, and homogeneity is appropriate, as these metrics are well-known for their efficacy in distinguishing different textural patterns inherent in various land covers. The experimental setup utilized a reasonable dataset split of 130 training and 111 test images. A notable finding is the critical influence of the K parameter in the KNN algorithm, with the study pinpointing K=7 as the optimal value yielding an impressive classification accuracy of 97.30%. The subsequent evaluation using a confusion matrix further solidifies the robustness and discriminative power of the proposed method in accurately separating the two distinct land cover types. The demonstrated high accuracy and efficiency of the combined GLCM and KNN approach underscore its significant potential for real-world application. This research offers a robust solution for automated mapping and monitoring systems, which could dramatically reduce the manual effort and time typically associated with land cover classification in agricultural and plantation settings. While the study focuses on a specific binary classification problem (rice fields vs. oil palms), the inherent modularity and proven performance suggest that this framework could be extended or adapted for classifying a wider array of land cover types. This work contributes a valuable and practical method to the field of remote sensing and image analysis, paving the way for more automated and data-driven decision-making in agricultural land management.

Full Text

Comments

You need to be logged in to post a comment.