Cross-domain prototype similarity correction for few-shot radar modulation signal recognition

Signal Processing, Год журнала: 2024, Номер 223, С. 109575 - 109575

Опубликована: Июнь 14, 2024

Язык: Английский

---

Multiscale Spatial–Spectral Dense Residual Attention Fusion Network for Spectral Reconstruction from Multispectral Images

Remote Sensing, Год журнала: 2025, Номер 17(3), С. 456 - 456

Опубликована: Янв. 29, 2025

Spectral reconstruction (SR) from multispectral images (MSIs) is a crucial task in remote sensing image processing, aiming to enhance the spectral resolution of MSIs produce hyperspectral (HSIs). However, most existing deep learning-based SR methods primarily focus on deeper network architectures, often overlooking importance extracting multiscale spatial and features MSIs. To bridge this gap, paper proposes spatial–spectral dense residual attention fusion (MS2Net) for SR. Specifically, considering nature land-cover types MSIs, three-dimensional hierarchical module designed embedded head proposed MS2Net extract features. Subsequently, we employ two-pathway architecture Both pathways are constructed with single-shot efficient feature learning composite soft salient Finally, extracted different integrated using an adaptive weighted reconstruct HSIs. Extensive experiments both simulated real-world datasets demonstrate that achieves superior performance compared state-of-the-art methods. Moreover, classification reconstructed HSIs show MS2Net-reconstructed achieve accuracy comparable real

Язык: Английский

---

A two-branch multiscale spectral-spatial feature extraction network for hyperspectral image classification

Journal of Information and Intelligence, Год журнала: 2024, Номер 2(3), С. 224 - 235

Опубликована: Март 9, 2024

In the field of hyperspectral image (HSI) classification in remote sensing, combination spectral and spatial features has gained considerable attention. addition, multiscale feature extraction approach is very effective at improving accuracy for HSIs, capable capturing a large amount intrinsic information. However, some existing methods extracting can only generate low-level consider limited scales, leading to low results, dense-connection based enhance propagation cost high model complexity. This paper presents Two-Branch Multiscale Spectral-Spatial Feature Extraction Network (TBMSSN) HSI Classification. We design Spectral (MSEFE) Spatial (MSAFE) modules improve representation, attention mechanism applied MSAFE module reduce redundant information representation multiscale. Then we densely connect series MSEFE or respectively two-branch framework balance efficiency effectiveness, alleviate vanishing-gradient problem strengthen propagation. To evaluate effectiveness proposed method, experimental results were carried out on bench mark datasets, demonstrating that TBMSSN obtained higher compared with several state-of-the-art methods.

Язык: Английский

---

Pyramid Cascaded Convolutional Neural Network with Graph Convolution for Hyperspectral Image Classification

Remote Sensing, Год журнала: 2024, Номер 16(16), С. 2942 - 2942

Опубликована: Авг. 11, 2024

Convolutional neural networks (CNNs) and graph convolutional (GCNs) have made considerable advances in hyperspectral image (HSI) classification. However, most CNN-based methods learn features at a single-scale HSI data, which may be insufficient for multi-scale feature extraction complex data scenes. To the relations among samples non-grid GCNs are employed combined with CNNs to process HSIs. Nevertheless, based on CNN-GCN overlook integration of pixel-wise spectral signatures. In this paper, we propose pyramid cascaded network convolution (PCCGC) It mainly comprises GCN-based subnetworks. Specifically, subnetwork, residual module extract multiscale spatial separately, can enhance robustness proposed model. Furthermore, an adaptive feature-weighted fusion strategy is utilized adaptively fuse features. band selection (BSNet) used signatures using nonlinear inter-band dependencies. Then, spectral-enhanced GCN important matrix. Subsequently, mutual-cooperative attention mechanism constructed align between BSNet-based matrix signature integration. Abundant experiments performed four widely real datasets show that our model achieves higher classification accuracy than fourteen other comparative methods, shows superior performance PCCGC over state-of-the-art methods.

Язык: Английский

---

TransHSI: A Hybrid CNN-Transformer Method for Disjoint Sample-Based Hyperspectral Image Classification

Remote Sensing, Год журнала: 2023, Номер 15(22), С. 5331 - 5331

Опубликована: Ноя. 12, 2023

Hyperspectral images’ (HSIs) classification research has seen significant progress with the use of convolutional neural networks (CNNs) and Transformer blocks. However, these studies primarily incorporated blocks at end their network architectures. Due to differences between spectral spatial features in HSIs, extraction both global local spectral–spatial remains incomplete. To address this challenge, paper introduces a novel method called TransHSI. This incorporates new feature module that leverages 3D CNNs fuse extract then combining 2D capture HSIs comprehensively. Furthermore, fusion is proposed, which not only integrates learned shallow deep but also applies semantic tokenizer transform fused features, enhancing discriminative power features. conducts experiments on three public datasets: Indian Pines, Pavia University, Data Fusion Contest 2018. The training test sets are selected based disjoint sampling strategy. We perform comparative analysis 11 traditional advanced HSI algorithms. experimental results demonstrate proposed method, TransHSI algorithm, achieves highest overall accuracies kappa coefficients, indicating competitive performance.

Язык: Английский

---

An Efficient and Adaptive Reconstructive Homogeneous Block-Based Local Tensor Robust PCA for Feature Extraction of Hyperspectral Images

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Год журнала: 2024, Номер 17, С. 4392 - 4407

Опубликована: Янв. 1, 2024

Model-driven tensor robust principal component analysis (TRPCA) has been widely applied to feature extraction of hyperspectral images (HSIs) and successfully protected 2-D spectral contextual information. Nevertheless, the current TRPCA-based methods still destroy underlying spatial-spectral joint features. Moreover, these global iterative algorithms commonly ignore heterogeneity different real-world regions, increase calculation burden, improve practice operating time. To solve issues, an efficient reconstructive homogeneous block-based local TRPCA is proposed for low-rank extraction, composed a block rebuilder extractor. The novel data-model-driven algorithm depending on data regulation. It remains primary spatial information extracts homogeneity characteristics spatial, spectral, variables, which provides more essential features further research than other model-driven models. Furthermore, our extractor elaborate divide-and-rule model that executes each extract adaptively, remarkably decreasing computing cost Experimental results six datasets demonstrate adaptive HSIs outperforms state-of-the-art algorithms.

Язык: Английский

---

IDACC: Image Dehazing Avoiding Color Cast Using a Novel Atmospheric Scattering Model

IEEE Access, Год журнала: 2024, Номер 12, С. 70160 - 70169

Опубликована: Янв. 1, 2024

Язык: Английский

---

Hyperspectral Target Detection Methods Based on Statistical Information: The Key Problems and the Corresponding Strategies

Remote Sensing, Год журнала: 2023, Номер 15(15), С. 3835 - 3835

Опубликована: Авг. 1, 2023

Target detection is an important area in the applications of hyperspectral remote sensing. Due to full use information target and background, algorithms based on statistical characteristics image are always occupy a dominant position field detection. From perspective information, we firstly presented detailed discussions key factors affecting results, including data origin, size, spectral variability target, number bands. Further, gave corresponding strategies for several common situations practical applications.

Язык: Английский

---

A deep learning ICDNET architecture for efficient classification of histopathological cancer cells using Gaussian noise images

Alexandria Engineering Journal, Год журнала: 2024, Номер 112, С. 37 - 48

Опубликована: Ноя. 1, 2024

Язык: Английский

---

Multi-scale nonlinear edge-based three-phase model for unsupervised hyperspectral feature extraction

Journal of Applied Remote Sensing, Год журнала: 2023, Номер 17(03)

Опубликована: Сен. 26, 2023

Unsupervised feature extraction techniques of hyperspectral images (HSIs) have recently drawn significant attention for their excellent performance and efficiency in classification. In some existing methods, the denoising process that reduces influence inherent noise is ignored, nonlinear edge characteristics multi-scale features help to classify still need be fully considered. To solve these issues, we employ a edge-based unsupervised three-phase model (UTPM) extraction. Specifically, initial phase, noise-adjusted principal components technique adopted lower improve proposed model. Then, neighbor band grouping designed reduce redundancy computational cost with information entropy. Because entropy can concretely reflect importance different bands same group, inner structure maximally preserved. Finally, utilize fusion on kernel low-rank entropic analysis extract combine it convolution algorithm fuse elements multiple scales classification performance. Compared several other classical or progressive algorithms, results three public HSI datasets validate effectiveness UTPM.

Язык: Английский

---