Distance Constraint-Based Generative Adversarial Networks for Hyperspectral Image Classification

Wang, R.; Zhao, Yue

doi:10.1109/tgrs.2023.3274778

Cited by 17 publications

(3 citation statements)

References 47 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Numerous models based on deep learning were proposed for HSI classification tasks. Some of the well-known baseline networks include recurrent neural networks [22], graph convolutional neural networks [23], autoencoders [24], generative adversarial networks [25], capsule networks [26], long short-term memory networks [27], and convolutional neural networks (CNNs) [28]. Among these methods, CNNs are the most widely used and can be categorized into 1D convolutional neural network (1D-CNN) [29], 2D convolutional neural network (2D-CNN) [30], and 3D convolutional neural network (3D-CNN) [31] based on their dimensions.…”

Section: Introductionmentioning

confidence: 99%

Multi-Level Feature Extraction Networks for Hyperspectral Image Classification

Fang,

Li,

Tian

et al. 2024

Remote Sensing

View full text Add to dashboard Cite

Hyperspectral image (HSI) classification plays a key role in the field of earth observation missions. Recently, transformer-based approaches have been widely used for HSI classification due to their ability to model long-range sequences. However, these methods face two main challenges. First, they treat HSI as linear vectors, disregarding their 3D attributes and spatial structure. Second, the repeated concatenation of encoders leads to information loss and gradient vanishing. To overcome these challenges, we propose a new solution called the multi-level feature extraction network (MLFEN). MLFEN consists of two sub-networks: the hybrid convolutional attention module (HCAM) and the enhanced dense vision transformer (EDVT). HCAM incorporates a band shift strategy to eliminate the edge effect of convolution and utilizes hybrid convolutional blocks to capture the 3D properties and spatial structure of HSI. Additionally, an attention module is introduced to identify strongly discriminative features. EDVT reconfigures the organization of original encoders by incorporating dense connections and adaptive feature fusion components, enabling faster propagation of information and mitigating the problem of gradient vanishing. Furthermore, we propose a novel sparse loss function to better fit the data distribution. Extensive experiments conducted on three public datasets demonstrate the significant advancements achieved by MLFEN.

show abstract

Section: Introductionmentioning

confidence: 99%

Multi-Level Feature Extraction Networks for Hyperspectral Image Classification

Fang,

Li,

Tian

et al. 2024

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…At present, a series of deep learning methods based on generative adversarial networks (GANs) have been successfully applied in image classification and recognition 23 , 24 . GANs were first introduced by Goodfellow 25 .…”

Section: Introductionmentioning

confidence: 99%

Semisupervised hyperspectral image classification based on generative adversarial networks and spectral angle distance

Zhan,

Wang,

2023

Sci Rep

View full text Add to dashboard Cite

Collecting ground truth labels for hyperspectral image classification is difficult and time-consuming. Without an adequate number of training samples, hyperspectral image (HSI) classification is a challenging problem. Using generative adversarial networks (GANs) is a promising technique for solving this problem because GANs can learn features from both labeled and unlabeled samples. The cost functions widely used in current GAN methods are suitable for 2D nature images. Compared with natural images, HSIs have a simpler one-dimensional structure that facilitates image generation. Motivated by the one-dimensional spectral features of HSIs, we propose a novel semisupervised algorithm for HSI classification by introducing spectral angle distance (SAD) as a loss function and employing multilayer feature fusion. Since the differences between spectra can be quickly calculated using the spectral angle distance, the convergence speed of the GAN can be improved, and the samples generated by the generator model in the GAN are closer to the real spectrum. Once the entire GAN model has been trained, the discriminator can extract multiscale features of labeled and unlabeled samples. The classifier is then trained for HSI classification using the multilayer features extracted from a few labeled samples by the discriminator. The proposed method was validated on four hyperspectral datasets: Pavia University, Indiana Pines, Salinas, and Tianshan. The experimental results show that the proposed model provides very promising results compared with other related state-of-the-art methods.

show abstract

“…These machine learning approaches for image classification can be subdivided into supervised (Aravind et al, 2018;Sheykhmousa et al, 2020) and unsupervised methods (Chen et al, 2018;Xie et al, 2018), classical optimization techniques (Meng et al, 2020), and stochastic optimization methods (Ahilan et al, 2019;Miao and Yang, 2021). However, deep learning models achieve higher classification accuracies as they exploit the spatial and spectral properties of the images, such as convolutional neural networks (CNN) (Chen et al, 2019;Feng et al, 2019), multimodal deep learning (Hong et al, 2021), stacked autoencoders (Zabalza et al, 2016;Su et al, 2018;Shi and Pun, 2020), recurrent neural networks (RNN) (Hang et al, 2019;Liang et al, 2022;Zhou et al, 2023), and generative adversarial networks (Shi et al, 2022;Qin et al, 2023). Another approach is based on constructing a graph (Ding et al, 2021;Yang et al, 2021), which depicts spatial and spectral relations for each pixel with their surroundings using an adjacency matrix; this enables a meaningful representation providing higher accuracy with less data for training the algorithms; nevertheless, the computational effort is increased.…”

Section: Introductionmentioning

confidence: 99%

A blind convolutional deep autoencoder for spectral unmixing of hyperspectral images over waterbodies

Alfaro-Mejía,

Manian,

Ortiz

et al. 2023

Front. Earth Sci.

View full text Add to dashboard Cite

Harmful algal blooms have dangerous repercussions for biodiversity, the ecosystem, and public health. Automatic identification based on remote sensing hyperspectral image analysis provides a valuable mechanism for extracting the spectral signatures of harmful algal blooms and their respective percentage in a region of interest. This paper proposes a new model called a non-symmetrical autoencoder for spectral unmixing to perform endmember extraction and fractional abundance estimation. The model is assessed in benchmark datasets, such as Jasper Ridge and Samson. Additionally, a case study of the HSI2 image acquired by NASA over Lake Erie in 2017 is conducted for extracting optical water types. The results using the proposed model for the benchmark datasets improve unmixing performance, as indicated by the spectral angle distance compared to five baseline algorithms. Improved results were obtained for various metrics. In the Samson dataset, the proposed model outperformed other methods for water (0.060) and soil (0.025) endmember extraction. Moreover, the proposed method exhibited superior performance in terms of mean spectral angle distance compared to the other five baseline algorithms. The non-symmetrical autoencoder for the spectral unmixing approach achieved better results for abundance map estimation, with a root mean square error of 0.091 for water and 0.187 for soil, compared to the ground truth. For the Jasper Ridge dataset, the non-symmetrical autoencoder for the spectral unmixing model excelled in the tree (0.039) and road (0.068) endmember extraction and also demonstrated improved results for water abundance maps (0.1121). The proposed model can identify the presence of chlorophyll-a in waterbodies. Chlorophyll-a is an essential indicator of the presence of the different concentrations of macrophytes and cyanobacteria. The non-symmetrical autoencoder for spectral unmixing achieves a value of 0.307 for the spectral angle distance metric compared to a reference ground truth spectral signature of chlorophyll-a. The source code for the proposed model, as implemented in this manuscript, can be found at https://github.com/EstefaniaAlfaro/autoencoder_owt_spectral.git.

show abstract

Distance Constraint-Based Generative Adversarial Networks for Hyperspectral Image Classification

Cited by 17 publications

References 47 publications

Multi-Level Feature Extraction Networks for Hyperspectral Image Classification

Multi-Level Feature Extraction Networks for Hyperspectral Image Classification

Semisupervised hyperspectral image classification based on generative adversarial networks and spectral angle distance

A blind convolutional deep autoencoder for spectral unmixing of hyperspectral images over waterbodies

Contact Info

Product

Resources

About