Semi-Supervised Classification of Hyperspectral Images Based on Extended Label Propagation and Rolling Guidance Filtering

Cui, Binge; Xie, Xiaoyun; Hao, Siyuan; Cui, Jiandi; Lu, Yan

doi:10.3390/rs10040515

Cited by 32 publications

(21 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…(1) According to Equation (11) and (12), solve the sparse representation coefficientα of each sample I i based on the reflectance componentR I i and the overcomplete dictionaryÃ. (2) According to Equation (13), calculate the information entropy of eachR I i to discriminating the purity of each sample I i .…”

Section: : For Eachmentioning

confidence: 99%

See 1 more Smart Citation

A Sparse Representation-Based Sample Pseudo-Labeling Method for Hyperspectral Image Classification

Cui

et al. 2020

Remote Sensing

Self Cite

View full text Add to dashboard Cite

Hyperspectral image classification methods may not achieve good performance when a limited number of training samples are provided. However, labeling sufficient samples of hyperspectral images to achieve adequate training is quite expensive and difficult. In this paper, we propose a novel sample pseudo-labeling method based on sparse representation (SRSPL) for hyperspectral image classification, in which sparse representation is used to select the purest samples to extend the training set. The proposed method consists of the following three steps. First, intrinsic image decomposition is used to obtain the reflectance components of hyperspectral images. Second, hyperspectral pixels are sparsely represented using an overcomplete dictionary composed of all training samples. Finally, information entropy is defined for the vectorized sparse representation, and then the pixels with low information entropy are selected as pseudo-labeled samples to augment the training set. The quality of the generated pseudo-labeled samples is evaluated based on classification accuracy, i.e., overall accuracy, average accuracy, and Kappa coefficient. Experimental results on four real hyperspectral data sets demonstrate excellent classification performance using the new added pseudo-labeled samples, which indicates that the generated samples are of high confidence.

show abstract

Section: : For Eachmentioning

confidence: 99%

“…To alleviate the "Hughes phenomenon", dimensionality reduction [8,9] and semi-supervised classification [10,11] have been extensively studied. The former can reduce the dimensions of hyperspectral images, and the latter can increase the number of training samples.…”

Section: Introductionmentioning

confidence: 99%

A Sparse Representation-Based Sample Pseudo-Labeling Method for Hyperspectral Image Classification

Cui

et al. 2020

Remote Sensing

Self Cite

View full text Add to dashboard Cite

show abstract

“…This could be not enough for a DCNN model to achieve effectively classification by extracting spatial features, and it is easy to cause overfitting. Therefore, the common hyperspectral CNN model [21,22] has only two convolution layers; thus, it is difficult to learn the combination features over a long distance. c) The number of parameters has to be kept as small as possible to achieve fast computational speed.…”

Section: Introductionmentioning

confidence: 99%

A hyperspectral image classification algorithm based on atrous convolution

Zhang

Zheng

Liu

et al. 2019

J Wireless Com Network

View full text Add to dashboard Cite

Hyperspectral images not only have high spectral dimension, but the spatial size of datasets containing such kind of images is also small. Aiming at this problem, we design the NG-APC (non-gridding multi-level concatenated Atrous Pyramid Convolution) module based on the combined atrous convolution. By expanding the receptive field of three layers convolution from 7 to 45, the module can obtain a distanced combination of the spectral features of hyperspectral pixels and solve the gridding problem of atrous convolution. In NG-APC module, we construct a 15layer Deep Convolutional Neural Networks (DCNN) model to classify each hyperspectral pixel. Through the experiments on the Pavia University dataset, the model reaches 97.9% accuracy while the parameter amount is only 0.25 M. Compared with other CNN algorithms, our method gets the best OA (Over All Accuracy) and Kappa metrics, at the same time, NG-APC module keeps good performance and high efficiency with smaller number of parameters.

show abstract

“…In [23], the weight support vector machine was used to keep the training effort low with a manually-collected set of pixels of the class of interest and a random sample of pixels. In [24], extended label propagation and rolling guidance filtering that uses superpixel propagation were applied to assign the same labels to all pixels within the superpixels that are generated by the image segmentation method.…”

Section: Introductionmentioning

confidence: 99%

Semi-Supervised Classification and Landscape Metrics for Mapping and Spatial Pattern Change Analysis of Tropical Forest Types in Thua Thien Hue Province, Vietnam

Tuong

Tani

Wang

et al. 2019

Forests

View full text Add to dashboard Cite

Research Highlights: In this study, we classified natural forest into four forest types using time-series multi-source remotely sensed data through a proposed semi-supervised model developed and validated for mapping forest types and assessing forest transition in Vietnam. Background and Objectives: Data on current forest state and changes detection are always essential for forest management and planning. There is, therefore, a need for improved tools to classify and evaluate forest dynamics more accurately and effectively. Our objective is to develop such tools using a semi-supervised model and landscape metrics to classify and map changes in natural forest types by using multi-source remotely sensed data. Materials and Methods: A combination of Landsat data with PALSAR and PALSAR-2 was used for forest classification through the proposed semi-supervised model. This model turned a kernel least square into a self-learning algorithm, trained by a small number of samples with given labels, and then used this classifier to assign labels to the unlabeled data. The overall accuracy, kappa, user’s accuracy, and producer’s accuracy were used to evaluate the classification accuracy by comparing the classified image with the results of ground truth interpretation. Based on the classified images, forest transition was evaluated using certain landscape metrics at the class and landscape levels. Results: The multi-source data approach achieved improved discrimination of forest types compared to only using single data (optical or radar data). Good classification accuracies were obtained, with kappas of 0.81, 0.76, and 0.74 for the years 2007, 2010, and 2016, respectively. The analysis of landscape metrics indicated that there were different behaviors in the four forest types, as well as provided much information about the trends in spatial pattern changes. Conclusions: This study highlights the utilization of a semi-supervised model in forest classification, and the analysis of forest transition using landscape metrics. However, future research should include a comparison of different models to estimate the improvement of the proposed model. Another important study that should be conducted is to test the proposed method on larger areas.

show abstract

Semi-Supervised Classification of Hyperspectral Images Based on Extended Label Propagation and Rolling Guidance Filtering

Cited by 32 publications

References 37 publications

A Sparse Representation-Based Sample Pseudo-Labeling Method for Hyperspectral Image Classification

A Sparse Representation-Based Sample Pseudo-Labeling Method for Hyperspectral Image Classification

A hyperspectral image classification algorithm based on atrous convolution

Semi-Supervised Classification and Landscape Metrics for Mapping and Spatial Pattern Change Analysis of Tropical Forest Types in Thua Thien Hue Province, Vietnam

Contact Info

Product

Resources

About