Unsupervised Bayesian Classification of a Hyperspectral Image Based on the Spectral Mixture Model and Markov Random Field

Fang, Yuan; Xu, Linlin; Peng, Junhuan; Yang, Honglei; Wong, Alexander; Clausi, David A.

doi:10.1109/jstars.2018.2858008

Cited by 20 publications

(9 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For k-means it is O(cdnt) because the distance from each pixel to each (original) mean is calculated as well as the new means. For the gaussian mixture model discussed in [42], it is O(cn 2 t) because of the energy minimization sub-algorithm which is used. For the SOM, we showed above the training can be performed on the ground, so the execution time complexity is more important.…”

Section: Discussionmentioning

confidence: 99%

“…Because it depends on the determination of the BMU which involves evaluation the distance between each pixel and all the nodes, the time complexity of evaluation it O(nz 2 d). Note when algorithms require spatial information, as [42] does, the operational procedures for moving training to the ground no longer work. The most computationally expensive step of the spectral clustering methods is the computation of the affinity matrix, which considers the similarity between the anchors and every pixel in the dataset, giving a complexity of O(dnm).…”

Section: Discussionmentioning

confidence: 99%

“…It has been used to cluster hyperspectral images and as preprocessing for classification [40,41]. An extension of this algorithm for hyperspectral data, called a Label-dependent Spectral Mixture Model with Markov Random Field (LSMM-MRF), is proposed in [42].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Self-Organizing Maps for Clustering Hyperspectral Images On-Board a CubeSat

2021

View full text Add to dashboard Cite

Hyperspectral remote sensing reveals detailed information about the optical response of a scene. Self-Organizing Maps (SOMs) can partition a hyperspectral dataset into clusters, both to enable more analysis on-board the imaging platform and to reduce downlink time. Here, the expected on-board performance of the SOM algorithm is calculated within two different satellite operational procedures: one in which the SOM is trained prior to imaging, and another in which the training is part of the operations. The two procedures are found to have advantages that are suitable to quite different situations. The computational requirements for SOMs of different sizes are benchmarked on the target hardware for the HYPSO-1 mission, and dimensionality reduction (DR) is tested as a way of reducing the SOM network size. We find that SOMs can run on the target on-board processing hardware, can be trained reasonably well using less than 0.1% of the total pixels in a scene, are accelerated by DR, and can achieve a relative quantization error of about 1% on scenes acquired by a previous hyperspectral imaging satellite, HICO. Moreover, if class labels are assigned to the nodes of the SOM, these networks can classify with a comparable accuracy to support vector machines, a common benchmark, on a few simple scenes.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Self-Organizing Maps for Clustering Hyperspectral Images On-Board a CubeSat

2021

View full text Add to dashboard Cite

show abstract

“…In Remote Sensing, the first applications of MRF occurred in late 1980's (Geman, Geman, 1984, Kittler, Föglein, 1984, Mohn et al, 1987. Recently it is still used for classification (Zhang et al, 2018a, Zhang et al, 2018b, Fang et al, 2018, Borhani, Ghassemian, 2014 and image segmentation. The main disadvantage of MRF is the need to explicitly model the distribution of the likelihood and assume that the observed image data are conditionally independent, given the labels.…”

Section: Introductionmentioning

confidence: 99%

An Unsupervised Segmentation Method for Remote Sensing Imagery Based on Conditional Random Fields

Soares

Körting

Fonseca

et al. 2020

Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

Abstract. Segmentation is a fundamental problem in image processing and a common operation in Remote Sensing, which has been widely used especially in Geographic Object-Based Image Analysis (GEOBIA). In this paper, we propose a new unsupervised segmentation algorithm based on the Conditional Random Fields (CRF) theory. The method relies on two levels of information: (1) that comes from an unsupervised classification with Fuzzy C-Means algorithm; (2) the 8-connected neighbourhood of a pixel. The algorithm was tested on a WorldView-2 multispectral image, with 2 m of spatial resolution. Results were evaluated using 6 quality measures, and their performance was compared with other image segmentation algorithms that are usually applied by the Remote Sensing community. Results indicate that the proposed algorithm achieved superior overall performance when compared others, despite some over-segmentation.

show abstract

“…The latter identifies the differences between image classes and finds the optimal classification surface between them, according to the conditional probability distribution of image features and classes. The typical methods based on production model include Bayesian classification model [4] and Gaussian mixture model [5], and the most representative methods based on discriminant model include nearest neighbor method [6], decision tree [7] and artificial neural network (ANN) [8]. Both types of image classification methods face the defects of being shallow classification techniques.…”

Section: Introductionmentioning

confidence: 99%

A Fast Recognition Algorithm of Online Social Network Images Based on Deep Learning

Meng¹,

Mao²,

Zhang³

et al. 2019

View full text Add to dashboard Cite

In recent years, a massive number of images have generated on the online social network (OSN). This calls for an efficient and rapid way to extract the information from the OSN images. This paper puts forward an OSN image classification method based on improved deep belief network (DBN) and support vector machine (SVM). In the proposed method, the image classification is enhanced by improving the self-adaptive learning rate based on incremental discrimination of reconstruction error and the weight update criteria with increasing momentum. The effectiveness of our method was confirmed through an image recognition experiment on OSN images obtained from Sina Weibo public platform, in comparison with four commonly used classification methods. The research results shed new light on feature extraction and classification of OSN images.

show abstract

Unsupervised Bayesian Classification of a Hyperspectral Image Based on the Spectral Mixture Model and Markov Random Field

Cited by 20 publications

References 35 publications

Self-Organizing Maps for Clustering Hyperspectral Images On-Board a CubeSat

Self-Organizing Maps for Clustering Hyperspectral Images On-Board a CubeSat

An Unsupervised Segmentation Method for Remote Sensing Imagery Based on Conditional Random Fields

A Fast Recognition Algorithm of Online Social Network Images Based on Deep Learning

Contact Info

Product

Resources

About