آمار
| تعداد نشریات | 45 |
| تعداد شماره‌ها | 1,550 |
| تعداد مقالات | 18,890 |
| تعداد مشاهده مقاله | 63,336,189 |
| تعداد دریافت فایل اصل مقاله | 23,089,864 |
Unsupervised Sentinel-2 Alteration Mapping via β-VAE, Attribute Profiles, and HDBSCAN: Gossan Detection and Albedo Dominance Diagnosis, Althearthar, Iraq | ||
| نشریه کاربرد سنجش از دور و سیستم اطلاعات جغرافیایی در علوم محیطی | ||
| مقالات آماده انتشار، پذیرفته شده، انتشار آنلاین از تاریخ 04 شهریور 1405 | ||
| نوع مقاله: مقاله پژوهشی | ||
| شناسه دیجیتال (DOI): 10.22034/rsgi.2026.73783.1171 | ||
| نویسندگان | ||
| روان خالد قاسم1؛ مریم بیاتی خطیبی* 2؛ مهرداد جیهونی1 | ||
| 1گروه سنجش از دور و GIS، دانشگاه تبریز،تبریز،ایران | ||
| 2'گروه سنجش از دور و GIS، دانشکده برنامه ریزی و علوم محیطی، دانشگاه تبریز. تبریز ایران. | ||
| چکیده | ||
| Objective: Porphyry-copper exploration requires hydrothermal alteration mapping, but remote-sensing workflows are limited by linear spectral assumptions, neglect of spatial context, and dependence on unavailable training labels. We present an unsupervised framework integrating a β-Variational Autoencoder (β-VAE), morphological attribute profiles (APs), and HDBSCAN to map alteration and iron-bearing phases from Sentinel-2 imagery without ground truth. Methods: A Sentinel-2A scene over the Althearthar porphyry system (Iraq) was masked for vegetation and water. Six bands were standardized and input to a β-VAE (β=4, latent dimension 8) trained on spatially blocked splits. Extended APs on the latent space produced features that were reduced via PCA; HDBSCAN clustering was validated using spectral shape, brightness ranking, and similarity matrices. Results: Six clusters and a noise class (~12%) were identified. Albedo dominated the clustering—five clusters were flat, brightness-driven, and over-partitioned background lithology by illumination. Crucially, Class 3 (8.7% of pixels) displayed a distinct iron-rich spectrum (elevated VNIR, SWIR absorption, strong NIR–SWIR contrast) consistent with supergene hematite/goethite, marking a validated iron-oxide gossan target. Conclusions: The β-VAE disentangled albedo, band-ratio, and slope dimensions, but the AP step re-amplified brightness because topography remained uncorrected. Albedo dominance arose from Sentinel-2's broad bands, high relief, and missing illumination correction. Despite this, Class 3 provides a direct exploration target and the noise class avoided forced classification of transitional zones. Proposed refinements include topographic correction, band ratios as input, cluster merging, and spectral validation. Explicit albedo-dominance diagnostics proved essential, offering a template for unsupervised deep-learning alteration mapping and underscoring that illumination correction remains indispensable. | ||
| کلیدواژه‌ها | ||
| Iron-oxide gossan detection؛ Sentinel2؛ β-Variational Autoencoder؛ Morphological Attribute Profiles؛ HDBSCAN clustering؛ Albedo dominance diagnosis | ||
| مراجع | ||
|
| ||
|
آمار تعداد مشاهده مقاله: 75 |
||