Hyperspectral image classification via a random patches network

Xu, Yonghao; Du, Bo; Zhang, Fan; Zhang, Liangpei

doi:10.1016/j.isprsjprs.2018.05.014

Cited by 185 publications

(102 citation statements)

References 47 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For HSIs, their neighboring bands are highly correlated in the spectral domain [22]. Figure 2 plots the correlation coefficient curves of the Pavia university scene HSI, a remote-sensing HSI widely applied in classification [23]. The value on the x-axis is the index of the current band H i .…”

Section: Correlation Analysismentioning

confidence: 99%

Hyperspectral Image Super-Resolution by Deep Spatial-Spectral Exploitation

Zhao

2019

Remote Sensing

View full text Add to dashboard Cite

Limited by the existing imagery sensors, hyperspectral images are characterized by high spectral resolution but low spatial resolution. The super-resolution (SR) technique aiming at enhancing the spatial resolution of the input image is a hot topic in computer vision. In this paper, we present a hyperspectral image (HSI) SR method based on a deep information distillation network (IDN) and an intra-fusion operation. Specifically, bands are firstly selected by a certain distance and super-resolved by an IDN. The IDN employs distillation blocks to gradually extract abundant and efficient features for reconstructing the selected bands. Second, the unselected bands are obtained via spectral correlation, yielding a coarse high-resolution (HR) HSI. Finally, the spectral-interpolated coarse HR HSI is intra-fused with the input HSI to achieve a finer HR HSI, making further use of the spatial-spectral information these unselected bands convey. Different from most existing fusion-based HSI SR methods, the proposed intra-fusion operation does not require any auxiliary co-registered image as the input, which makes this method more practical. Moreover, contrary to most single-based HSI SR methods whose performance decreases significantly as the image quality gets worse, the proposal deeply utilizes the spatial-spectral information and the mapping knowledge provided by the IDN, which achieves more robust performance. Experimental data and comparative analysis have demonstrated the effectiveness of this method.

show abstract

Section: Correlation Analysismentioning

confidence: 99%

Hyperspectral Image Super-Resolution by Deep Spatial-Spectral Exploitation

Zhao

2019

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…G-RPNet same as RPNet has two major layers, including principal component analysis (PCA) and data whitening, and convolution with random patch selection and rectified linear units (ReLU). More details of these layers are as follows [15]:…”

Section: G-rpnetmentioning

confidence: 99%

“…Rotation based deep forest (RBDF) is proposed by Cao et al [14] in which the output probability of each layer is used as the supplement feature of the next layer, and rotation forest is used to increase the discriminative power of features. As a recent study, the random patches network (RPNet) is proposed by Xu et al [15] that simulates a deep model with the principal component analysis and convolutional kernels that are directly extracted from image random patches without any training stage.…”

Section: Introductionmentioning

confidence: 99%

Spatial-Spectral Random Patches Network for Classification of Hyperspectral Images

Beirami¹,

Mokhtarzade²

2019

View full text Add to dashboard Cite

Random patches network (RPNet) is an emerging deep learning method that can effectively extract the deep features from hyperspectral images. However, this network only relies on spectral bands in feature extraction, failing to make use of the information-rich spatial features. This paper puts forward another variant of the RPNet, named G-RPNet. The proposed network extracts the deep hierarchical Gabor features, with Gabor spatial features as inputs. The extracted deep hierarchical features were stacked to those extracted by the RPNet, and the final feature vectors were classified by the support vector machine (SVM). The integrated feature vectors inherit the merits of the deep hierarchical features of both RPNet and G-RPNet, laying a solid basis for the classification of hyperspectral images. Experiments were conducted on two real hyperspectral images (Indian Pines and Pavia University) from agricultural and urban areas. The results prove the superiority of the proposed method in the classification of hyperspectral images over some recent shallow and deep spatial-spectral classification techniques.

show abstract

“…Limited training samples in relation to high parameters of CNN was also studied in (Yu et al, 2017) by considering data augmentation, appropriate convolutional kernel size, larger drop rates in the dropout layers, discarding the most commonly used max-pooling layers and fully connected layers. Another DL-based classification method named RPNet (Random Patches Network) was designed by Xu et al, (2018). In their network, convolutional kernels are selected randomly from input patches without any training.…”

Section: Introductionmentioning

confidence: 99%

Hyperspectral Image Classification by Exploiting Convolutional Neural Networks

Hosseiny

Rastiveis

Daneshtalab

2019

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

View full text Add to dashboard Cite

High spectral dimensionality of hyperspectral images makes them useful data resources for earth observation in many remote sensing applications. In this case, the convolutional neural network (CNN) can help to extract deep and robust features from hyperspectral images. The main goal of this paper is to use deep learning concept to extract deep features from hyperspectral datasets to achieve better classification results. In this study, after pre-processing step, data is fed to a CNN in order to extract deep features. Extracted features are then imported in a multi-layer perceptron (MLP) network as our selected classifier. Obtained classification accuracies, based on training sample size, vary from 94.3 to 97.17% and 92.35 to 98.14% for Salinas and Pavia datasets, respectively. These results expressed more than 10% improvements compared to the classic MLP classification technique.

show abstract

Hyperspectral image classification via a random patches network

Cited by 185 publications

References 47 publications

Hyperspectral Image Super-Resolution by Deep Spatial-Spectral Exploitation

Hyperspectral Image Super-Resolution by Deep Spatial-Spectral Exploitation

Spatial-Spectral Random Patches Network for Classification of Hyperspectral Images

Hyperspectral Image Classification by Exploiting Convolutional Neural Networks

Contact Info

Product

Resources

About