Pixel classification of large-size hyperspectral images by affinity propagation

Abstract

Affinity propagation (AP) is now among the most used methods for unsupervised classification. However, it has two major drawbacks: (1) the number of classes (NCs) is over-estimated when the preference parameter value is initialized as the median value of the similarity matrix; and (2) the partitioning of large-size hyperspectral images is hampered by its quadratic computational complexity. To overcome these two drawbacks, we propose an approach which consists of reducing the number of pixels to be classified before the application of AP. To reduce the number of pixels, the hyperspectral image is divided into blocks, and the reduction step is then independently applied within each block. This step requires less memory storage, since the calculation of the full similarity matrix is no longer required. AP is applied on the new set of pixels, which is then set up from the representatives of each previously formed cluster and nonaggregated pixels. To correctly estimate the NCs, we introduced a bisection method which aims to assess intermediate classification results using a criterion based on pixel interclass variance. The application of this approach on hyperspectral images shows that our results are efficient and independent of the block size. © The Authors. Published by SPIE under a Creative Commons Attribution 3.0 Unported License. Distribution or reproduction of this work in whole or in part requires full attribution of the original publication, including its DOI.

Parallelizing Affinity Propagation Using Graphics Processing Units for Spatial Cluster Analysis over Big Geospatial Data

“…In geospatial applications, several prior works (Xia et al 2009; Yang et al 2010; Napoleon et al 2012; Chehdi et al 2014) were able to handle only tiny datasets as prototypes to test the AP approach, because the images mentioned in these publications have a dimension of only several dozens or hundreds of pixels. The image size allowed by the AP algorithm in the MATLAB environment cannot exceed 3,000 pixels.…”

“…The size of sample data provided online 1 ranges from 400 to 17,770 points, pixels, or units. A review of AP applications in remote sensing (Xia et al 2009; Yang et al 2010; Napoleon et al 2012; Chehdi et al 2014) reveals that the size of those sample images is less than 100-by-100 pixels. If AP cannot process large amounts of data, it can hardly be adopted in geospatial applications.…”

Data-driven process decomposition and robust online distributed modelling for large-scale processes

“…Affinity propagation (AP) is a new exemplarbased clustering method proposed by Frey andDueck in 2007 (Frey &Dueck, 2007). Compared with other methods, its distinguishing feature is that AP considers all the data points as potential exemplars and identifies clusters automatically (Ding, Ma, & Shi, 2013;Jia, Ding, Meng, & Fan, 2014), which can avoid many poor clustering solutions caused by unlucky initialisations and hard decisions (Chehdi & Soltani, 2008). Hence, AP cluster algorithm is utilised in our system decomposition process to classify the key controlled parameters and the result is the number of subsystems and corresponding outputs.…”

Stable and unsupervised fuzzy C-means method and its validation in the context of multicomponent images

“…The GT related to this image includes six different invasive and noninvasive vegetation classes, namely, Phragmites australis, Arundo donax, Tamarix, Ulmus minor, Pinus halepensis, and peach trees. 16 The GT comprises 9059 pixels. For each component of this image, FCMO is first applied using the feature set described above.…”