Hierarchical CNN Classification of Hyperspectral Images Based on 3-D Attention Soft Augmentation

Miao, Xinyuan; Zhang, Ye; Zhang, Junping; Liang, Xuejian

doi:10.1109/jstars.2022.3174301

Cited by 4 publications

(4 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Unlike self-augmentation, mutual augmentation of HSI samples works by exploiting several samples, e.g., mix-up, borrowed from machine learning literature [59,60], and consistency of labels of pixel and block that form pairs [58]. Along the lines of mix-up, Miao et al applied 3D CutMix and trainable spectral-spatial attention module as DA to guide the CNN classifier to attend to the discriminative features of HSI [61]. Wang et al proposed two strategies of weighting several nearest neighboring samples of the central sample as its augmentation, i.e., applying different sizes of patches and iteratively applying them to the new augmented image [62].…”

Section: Augment Samplesmentioning

confidence: 99%

Nearest Neighboring Self-Supervised Learning for Hyperspectral Image Classification

Qin

Zhao

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

Recently, state-of-the-art classification performance of natural images has been obtained by self-supervised learning (S2L) as it can generate latent features through learning between different views of the same images. However, the latent semantic information of similar images has hardly been exploited by these S2L-based methods. Consequently, to explore the potential of S2L between similar samples in hyperspectral image classification (HSIC), we propose the nearest neighboring self-supervised learning (N2SSL) method, by interacting between different augmentations of reliable nearest neighboring pairs (RN2Ps) of HSI samples in the framework of bootstrap your own latent (BYOL). Specifically, there are four main steps: pretraining of spectral spatial residual network (SSRN)-based BYOL, generation of nearest neighboring pairs (N2Ps), training of BYOL based on RN2P, final classification. Experimental results of three benchmark HSIs validated that S2L on similar samples can facilitate subsequent classification. Moreover, we found that BYOL trained on an un-related HSI can be fine-tuned for classification of other HSIs with less computational cost and higher accuracy than training from scratch. Beyond the methodology, we present a comprehensive review of HSI-related data augmentation (DA), which is meaningful to future research of S2L on HSIs.

show abstract

Section: Augment Samplesmentioning

confidence: 99%

Nearest Neighboring Self-Supervised Learning for Hyperspectral Image Classification

Qin

Zhao

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…These sets are obtained via simplified augmentation operations and can be applied to multimodal application scenarios. Similarly, work in [26], [27], [28], [29], and [30] proposes the use of a Convolutional Network with Twofold Feature Augmentations, Proto-MaxUp (PM), Conditional GAN, Hierarchical CNN with Soft Augmentation (HCNN SA), and Mask Region CNN, for estimation of high-density image sets under different application scenarios. These models can improve classification efficiency under multimodal scenarios.…”

Section: Brief Review Of Image Augmentation Modelsmentioning

confidence: 99%

“…Due to this, the model can showcase superior accuracy performance under large image sets. Similarly, the precision of classification [60] was evaluated via (28),…”

Section: Via (27)mentioning

confidence: 99%

MBAHIL: Design of a Multimodal Hybrid Bioinspired Model for Augmentation of Hyperspectral Imagery via Iterative Learning for Continuous Efficiency Enhancements

et al. 2023

View full text Add to dashboard Cite

The augmentation of hyperspectral images requires the design of high-density feature analysis & band-fusion models that can generate multimodal imagery from limited information sets. The feature analysis models use deep learning operations to maximize inter-class variance while minimizing inter-class variance levels for efficient classification operations. When combined with intelligent band-fusion methods, such models allow the augmentation model to enhance its classification efficiency under different use cases. Existing band-fusion-based augmentation models for hyperspectral images do not incorporate continuous efficiency enhancements and showcase higher complexity levels. Furthermore, these models can't be scaled for more varied use cases because their use is restricted to specific image types. To overcome these issues, we designed a novel multimodal hybrid bioinspired model for the augmentation of hyperspectral imagery via iterative learning for continuous efficiency enhancements. The proposed model initially represents input images into Fourier, Laplacian, Cosine, multimodal Wavelet, Mellin, and Z-Transform domains, which will assist in describing the images in multimodal domains. These transformed image sets are passed through a convolutional filter to extract windowed feature sets. A Grey Wolf Optimizer (GWO) is used to identify high inter-class variance features from the extracted image sets, which assists in selecting transformed images that can help improve hyperspectral classification performance. The selected hyperspectral images are fused via a Bacterial Foraging Optimization (BFO) model, which assists in reducing intra-class variance levels. The final set of selected images is intelligently augmented via Particle Swarm Optimization (PSO), which performs rotation, zooming, shifting, and brightness variation operations selectively. These augmented images are classified via a customized VGGNet-19-based Convolutional Neural Network (CNN) classifier that assists in continuously estimating accuracy levels for different application scenarios. Based on these accuracy levels, the model is reconfigured via hyperparameter tuning of GWO and PSO optimizers. Due to combining these models and incremental accuracy optimizations, the proposed model has improved its hyperspectral classification accuracy by 10.6% and precision by 10.4%, as compared to standard deep learning-based augmentation techniques.INDEX TERMS Bacterial foraging optimization, grey wolf optimizer, hyperspectral images, particle swarm optimization, VGGNet-19.The associate editor coordinating the review of this manuscript and approving it for publication was Mehul S. Raval .

show abstract

“…There are some shortcut connections in the residual learning network [21]. These connections skip some layers and pass the original data directly to subsequent layers [22]. When the extreme case is encountered that the new layer does not learn any data, the new layer in the residual learning network can directly copy the original data.…”

Section: Figure 3 the Framework Of Residual Learningmentioning

confidence: 99%

ODET: Optimized Deep ELM-based Transfer Learning for Breast Cancer Explainable Detection

Zhu¹,

Wang²

2022

EAI Endorsed Scal Inf Syst

View full text Add to dashboard Cite

INTRODUCTION: Breast cancer is one of the most common malignant tumors in women, and the incidence rate is increasing year by year. Women in every country in the world may develop breast cancer at any age after puberty. The cause of breast cancer is not fully understood. At present, the main methods of breast cancer detection are inefficient. Researchers are trying to use computer technology to detect breast cancer. But there are some still limitations. METHODS: We propose a network (ODET) to detect breast cancer based on ultrasound images. In this paper, we use ResNet50 as the backbone model. We make some modifications to the backbone model by deep ELM-based transfer learning. After these modifications, the network is named DET. However, DET still has some shortcomings because the parameters in DET are randomly assigned and will not change in the experiment. In this case, we select BA to optimize DET. The optimized DET is named ODET. RESULTS: The proposed ODET gets the F1-score (F1), precision (PRE), specificity (SPE), sensitivity (SEN), and accuracy (ACC) are 93.16%±1.12%, 93.28%±1.36%, 98.63%±0.31%, 93.96%±1.85%, and 97.84%±0.37%, respectively. CONCLUSION: It proves that the proposed ODET is an effective method for breast cancer detection.

show abstract

Hierarchical CNN Classification of Hyperspectral Images Based on 3-D Attention Soft Augmentation

Cited by 4 publications

References 45 publications

Nearest Neighboring Self-Supervised Learning for Hyperspectral Image Classification

Nearest Neighboring Self-Supervised Learning for Hyperspectral Image Classification

MBAHIL: Design of a Multimodal Hybrid Bioinspired Model for Augmentation of Hyperspectral Imagery via Iterative Learning for Continuous Efficiency Enhancements

ODET: Optimized Deep ELM-based Transfer Learning for Breast Cancer Explainable Detection

Contact Info

Product

Resources

About