Spectral clustering has been successfully used in various applications, thanks to its properties such as no requirement of a parametric model, ability to extract clusters of different characteristics and easy implementation. However, it is often infeasible for large datasets due to its heavy computational load and memory requirement. To utilize its advantages for large datasets, it is applied to the dataset representatives (either obtained by quantization or sampling) rather than the data samples, which is called approximate spectral clustering. This necessitates novel approaches for defining similarities based on representatives exploiting the data characteristics, in addition to the traditional Euclidean distance based similarities. To address this challenge, we propose similarity measures based on geodesic distances and local density distribution. Our experiments using datasets with varying cluster statistics show that the proposed geodesic based similarities are successful for approximate spectral clustering with high accuracies.
Unsupervised clustering of high spatial resolution remote-sensing images plays a significant role in detailed landcover identification, especially for agricultural and environmental monitoring. A recently promising method is approximate spectral clustering (SC) which enables spectral partitioning for large datasets to extract clusters with distinct characteristics without a parametric model. It also facilitates the use of various information types via advanced similarity criteria. However, it requires an empirical selection of a similarity criterion optimal for the corresponding application. To address this challenge, we propose an approximate SC ensemble (ASCE2) which fuses partitionings obtained by different similarity representations. Contrary to existing spectral ensembles for remote-sensing applications, the proposed ASCE2 employs neural gas quantization instead of random sampling, advanced similarity criteria instead of traditional distance-based Gaussian kernel with different decay parameters, and a two-level ensemble. We evaluate the proposed ASCE2 with three measures (accuracy, adjusted Rand index, and normalized mutual information) using five remote-sensing images, two of which are commonly available. We apply the ASCE2 in two applications for agricultural monitoring: 1) land-cover identification to determine orchard fields using a WorldView-2 image (0.5-m spatial resolution) and 2) finding lands in good agricultural condition using multitemporal RapidEye images (5-m spatial resolution). Experimental results indicate a significant betterment of the resulting partitionings obtained by the proposed ensemble, with respect to the evaluation measures in these applications.
Approximate spectral clustering (ASC), a recently popular approach for unsupervised land cover identification, applies spectral clustering on a reduced set of data representatives (found by sampling or quantization). ASC enables extraction of clusters with different characteristics by utilizing various information types (such as distance, local density distribution and data topology) for accurate similarity definition. However, selection of a sampling / quantization method and a similarity criterion is of great importance for optimal clustering. Alternatively, we propose sampling based ASC ensemble (SASCE) to exploit different similarity criteria with selective sampling by merging their partitionings into a consensus result. We show the outperformance of the proposed ensemble SASCE on four land cover datasets in comparison with their individual clusterings.
Özetçe -Fındık, ülkemizde ekonomik ve çevresel açıdan büyük önem taşıyan bir meyvedir. Bu yüzden fındık arazilerinin belirlenmesi, korunması, kontrol edilmesi ve yönetilmesi oldukça önemlidir. Uzaktan algılama teknolojisiyle bu bölgelerden elde edilen görüntüler sayesinde bu süreç daha hızlı ve dogru olarak gerçekleştirilir. Bu noktada en kritik aşama alınan görüntülerin dogru birşekilde (egitmensiz) degerlendirilebilmesidir. Bu açıdan öbekleme yöntemleri önemlidir. Bu yöntemlerden spektral öbekleme (SÖ) parametrik bir modele dayanmaması, farklı karakteristik özelliklere sahip öbekleri çıkartma becerisi sayesinde başarılı olmasına ragmen yüksek hesaplama yükü ve hafıza gereksinimi sebebiyle uzaktan algılama görüntülerinde dogrudan uygulanamamaktadır. Bunun yerine spektral öbeklemeyi örnekleme ya da nicemlemeyle seçilmiş veri temsilcilerine uygulayan yaklaşık spektral öbekleme (YSÖ) kullanılmaktadır. YSÖ, spektral öbeklemenin artılarından yararlanmasının yanısıra temsilciler sayesinde veri topolojisi, yerel yogunluk daglımı, Öklid veya jeodezik uzaklık gibi farklı tipte bilgileri kullanarak veri özelliklerini daha ayrıntılı yansıtan benzerlik ölçütlerinin kullanılmasını mümkün kılar. Çok çeşitli benzerlik ölçütlerinin ve kullanıldıgı örnekleme/nicemleme yaklaşımının farklı küme yapıları üzerinde degişen başarılara sahip olması yüzünden hangi ölçütün hangi veri azaltma yöntemiyle kullanılacagına karar vermek zordur. Bu zorluktan kurtulmak ve elde edilen tüm sonuçları bir araya getirerek en başarılı öbekleme sonucunu elde etmek amacıyla, bu çalışmada seçimli örnekleme tabanlı yaklaşık spektral öbekleme birleşimi (SÖYSÖB) sunulmuş ve fındık bahçelerinin bulunmasındaki başarısı gösterilmiştir.Anahtar Kelimeler-yaklaşık spektral öbekleme, jeodezik benzerlik, uzaktan algılama, fındık bahçeleri, oy çoklugu birleşimi.Abstract-Considering the economic and environmental aspects, hazelnut orchards are of great importance in Turkey. It is crucial to develop methods for detecting, monitoring, protecting and managing these orchards. This can be done exactly and fast by evaluating the remote sensing images of these areas. For this aim clustering methods are so popular due to their unsupervised nature. Particularly, spectral clustering can be useful thanks to its ability to extract clusters of different structures and its easy implementation. However, its direct use in remote sensing Bu çalışma kısmen 112E195 nolu TÜBİTAK Kariyer Projesiyle desteklenmektedir. Ayrıca K. Taşdemir, FP7 Marie Curier Career Integration Grant (IAM4MARS) ile de desteklenmektedir. images is infeasible due to its high computational and memory cost. That's why its indirect implementation, which is named approximate spectral clustering (ASC), is applied on quantized or sampled prototypes of the image. ASC not only takes advantages of spectral clustering but also utilizes different information types (such as topology and density) on the prototype level for effective similarity definition. Although ASC can be used with various similarity measures, the optimum...
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.