Fauzia Aristalindra, Abdul Basith

Penerapan SAM-Geo untuk Delineasi Otomatis Batas Bidang Tanah Pertanian pada Ortofoto

Introduction

Penerapan sam-geo untuk delineasi otomatis batas bidang tanah pertanian pada ortofoto. Evaluasi SAM-Geo untuk delineasi otomatis batas bidang tanah pertanian dari ortofoto di Indonesia. Solusi presisi untuk manajemen agraria efisien, meski tantangan di area heterogen.

Abstract

Accurate data on agricultural parcel boundaries are essential to support efficient and equitable agrarian management. Conventional methods such as terrestrial surveys and manual digitization are often costly, time-consuming, and inconsistent. Advances in artificial intelligence-based segmentation models, particularly the Segment Anything Model for Geospatial (SAM-Geo), offer new opportunities to accelerate the automatic delineation of agricultural land parcels. This study aims to evaluate the performance of SAM-Geo in extracting agricultural parcel boundaries from 3.98 cm resolution orthophotos in Sumberrahayu Village, Moyudan Subdistrict, Sleman Regency, Yogyakarta Special Region, Indonesia. The research process includes orthophoto preprocessing, SAM-Geo implementation, mask filtering, and accuracy assessment using both area-based metrics (precision, recall, F1-score, and IoU) and boundary-based metrics (boundary precision, recall, and F1-score) with a 1 m buffer tolerance. The results indicate that SAM-Geo can produce highly precise boundary delineation in homogeneous areas, achieving F1-score and IoU values above 96%, while performance declines in heterogeneous areas due to complex land cover conditions. Overall, this study provides one of the first empirical evaluations of SAM-Geo in agricultural landscapes in Indonesia and highlights its potential as an effective approach for agricultural parcel boundary mapping. Ketersediaan data batas bidang tanah pertanian yang akurat menjadi prasyarat penting dalam mendukung pengelolaan agraria yang efisien dan berkeadilan. Metode konvensional seperti survei terestris dan digitasi manual seringkali memerlukan biaya tinggi, waktu lama, serta menghasilkan ketidakkonsistenan data. Perkembangan model segmentasi berbasis kecerdasan buatan, khususnya Segment Anything Model for Geospatial (SAM-Geo), membuka peluang baru untuk mempercepat delineasi batas bidang tanah pertanian secara otomatis. Penelitian ini bertujuan mengevaluasi kinerja SAM-Geo dalam mengekstraksi batas bidang pertanian dari ortofoto beresolusi 3,98 cm di Kalurahan Sumberrahayu, Kapanewon Moyudan, Kabupaten Sleman, D.I. Yogyakarta. Metode penelitian mencakup pra-pemrosesan ortofoto, penerapan SAM-Geo, mask filtering, serta evaluasi akurasi menggunakan metrik berbasis area (precision, recall, F1-score, and IoU) dan berbasis batas (boundary precision, recall, and F1-score) dengan toleransi buffer 1 m. Hasil penelitian menunjukkan SAM-Geo menghasilkan delineasi batas sangat presisi pada area homogen dengan F1-score dan IoU di atas 96%, sedangkan performa menurun pada area heterogen akibat kompleksitas tutupan lahan. Temuan ini menegaskan potensi SAM-Geo sebagai pendekatan efektif untuk pemetaan batas bidang pertanian di Indonesia.

Review

The paper, "Penerapan SAM-Geo untuk Delineasi Otomatis Batas Bidang Tanah Pertanian pada Ortofoto," addresses a critical challenge in agrarian management: the accurate and efficient delineation of agricultural parcel boundaries. The authors rightly highlight the significant drawbacks of conventional methods, such as high costs, time consumption, and data inconsistencies, which hinder effective and equitable land administration. By exploring the capabilities of the Segment Anything Model for Geospatial (SAM-Geo), a cutting-edge AI-based segmentation model, this study positions itself at the forefront of leveraging advanced technology to overcome these long-standing issues, aiming to accelerate the automatic mapping of agricultural land parcels. The methodology employed is robust, focusing on evaluating SAM-Geo's performance in extracting parcel boundaries from high-resolution orthophotos (3.98 cm) in a specific Indonesian agricultural landscape. The research systematically covers orthophoto preprocessing, SAM-Geo implementation, and crucial mask filtering steps. A significant strength lies in its comprehensive accuracy assessment, utilizing both area-based metrics (precision, recall, F1-score, IoU) and boundary-based metrics (boundary precision, recall, and F1-score) with a practical 1-meter buffer tolerance. The findings reveal a nuanced performance: SAM-Geo demonstrates exceptional precision in homogeneous areas, achieving F1-score and IoU values exceeding 96%, underscoring its potential in simpler agricultural settings. However, its performance notably declines in heterogeneous areas, attributed to the inherent complexity of land cover conditions. This study makes a valuable contribution as one of the pioneering empirical evaluations of SAM-Geo within an Indonesian agricultural context, providing crucial insights into its real-world applicability. The findings strongly affirm SAM-Geo's potential as an effective and scalable approach for agricultural parcel boundary mapping, which can significantly enhance the efficiency and fairness of agrarian management. While the observed performance drop in heterogeneous landscapes presents a challenge, it also points to an important area for future research, perhaps involving integration with other data sources or advanced post-processing techniques. Overall, the research successfully demonstrates the significant promise of AI-driven geospatial models for solving practical land administration problems and lays a foundation for wider adoption and refinement of these technologies in similar contexts globally.

Full Text

Comments

You need to be logged in to post a comment.