Rotation-Invariant Attention Network for Hyperspectral Image Classification

IEEE Transactions on Image Processing, Journal Year: 2022, Volume and Issue: 31, P. 4251 - 4265

Published: Jan. 1, 2022

Hyperspectral image (HSI) classification refers to identifying land-cover categories of pixels based on spectral signatures and spatial information HSIs. In recent deep learning-based methods, explore the HSIs, HSI patch is usually cropped from original as input. And 3×3 convolution utilized a key component capture features for classification. However, sensitive rotation inputs, which results in that methods perform worse rotated To alleviate this problem, rotation-invariant attention network (RIAN) proposed First, center (CSpeA) module designed avoid influence other suppress redundant bands. Then, rectified (RSpaA) replace extracting spectral-spatial patches. The CSpeA module, 1×1 RSpaA are build RIAN Experimental demonstrate invariant HSIs has superior performance, e.g., achieving an overall accuracy 86.53% (1.04% improvement) Houston database.

Language: Английский

---

EMTCAL: Efficient Multiscale Transformer and Cross-Level Attention Learning for Remote Sensing Scene Classification

IEEE Transactions on Geoscience and Remote Sensing, Journal Year: 2022, Volume and Issue: 60, P. 1 - 15

Published: Jan. 1, 2022

In recent years, convolutional neural network (CNN)-based methods have been widely used for remote sensing (RS) scene classification tasks and achieved excellent results. However, CNNs are not good at exploring contextual information, which is essential fully understanding RS scenes. A new model named transformer attracts researchers' attention to address this problem, skilled in mining the latent information Nevertheless, since contents of scenes diverse type various scale, performance original cannot reach what we expect. addition, due specific self-attention mechanism, time costs high, hinders its practicability community. To overcome above limitations, propose a efficient multi-scale cross-level learning (EMTCAL) paper. EMTCAL combines advantages CNN mine within fully. First, it uses multi-layer feature extraction module (MFEM) acquire global visual features multi-level from Second, (CIEM) proposed capture rich features. CIEM, taking characteristics computational complexity into account, an (EMST). EMST can abundant knowledge with scales hidden their inherent relations small-time costs. Third, (CLAM) developed aggregate explore correlations Finally, class score fusion (CSFM) designed integrate contributions aggregated discriminative representations. Extensive experiments conducted on three public data sets. The positive results demonstrate that our achieve superior outperform many state-of-the-art methods.

Language: Английский

---

A Synergistical Attention Model for Semantic Segmentation of Remote Sensing Images

IEEE Transactions on Geoscience and Remote Sensing, Journal Year: 2023, Volume and Issue: 61, P. 1 - 16

Published: Jan. 1, 2023

In remotely sensed images, high intraclass variance and interclass similarity are ubiquitous due to complex scenes objects with multivariate features, making semantic segmentation a challenging task. Deep convolutional neural networks can solve this problem by modeling the context of features improving their discriminability. However, current learning paradigms model feature affinity in spatial dimension channel separately then fuse them sequential or parallel manner, leading suboptimal performance. study, we first analyze practically summarize it as attention bias that reduces capability network distinguishing weak discretely distributed from wide-range internal connectivity, when modeled only domain. To jointly both affinity, design synergistic module (SAM), which allows for channelwise extraction while preserving details. addition, propose perception (SAPNet) remote sensing images. The hierarchical-embedded aggregates SAM-refined decoded features. As result, SAPNet enriches inference clues desired Experiments on three benchmark datasets show is competitive accuracy adaptability compared state-of-the-art methods. experiments also validate hypothesis efficiency SAM.

Language: Английский

---

A Review of Practical AI for Remote Sensing in Earth Sciences

Remote Sensing, Journal Year: 2023, Volume and Issue: 15(16), P. 4112 - 4112

Published: Aug. 21, 2023

Integrating Artificial Intelligence (AI) techniques with remote sensing holds great potential for revolutionizing data analysis and applications in many domains of Earth sciences. This review paper synthesizes the existing literature on AI sensing, consolidating analyzing methodologies, outcomes, limitations. The primary objectives are to identify research gaps, assess effectiveness approaches practice, highlight emerging trends challenges. We explore diverse including image classification, land cover mapping, object detection, change hyperspectral radar analysis, fusion. present an overview technologies, methods employed, relevant use cases. further challenges associated practical such as quality availability, model uncertainty interpretability, integration domain expertise well solutions, advancements, future directions. provide a comprehensive researchers, practitioners, decision makers, informing at exciting intersection sensing.

Language: Английский

---

Classification via Structure-Preserved Hypergraph Convolution Network for Hyperspectral Image

IEEE Transactions on Geoscience and Remote Sensing, Journal Year: 2023, Volume and Issue: 61, P. 1 - 13

Published: Jan. 1, 2023

Graph convolutional network (GCN) as a combination of deep learning and graph has gained increasing attention in hyperspectral image (HSI) classification. However, most GCN methods consider the simple point-to-point structure between two pixels rather than high-order multiple pixels, which is contradict with real feature distribution ground object. And nonlinear property HSI also brings challenge for precise structural representation GCN. To tackle these problems, this work proposes preserved hypergraph convolution (SPHGCN). It first builds neighborhood reconstruction (MNR) model to reveal essential resemblance spectral space. With structure, SPHGCN designs operation irregular aggregation among similar from different regions, achieves more discriminative features pixel nodes. Meanwhile, preservation layer built optimize under guidance structure. Moreover, integrates local regular learn structured semantic HSI. This strategy breaks boundary restriction traditional aggregates across patches. Experiments on three data sets indicate that outperforms few state-of-the-art

Language: Английский

---

Abundance Matrix Correlation Analysis Network Based on Hierarchical Multihead Self-Cross-Hybrid Attention for Hyperspectral Change Detection

IEEE Transactions on Geoscience and Remote Sensing, Journal Year: 2023, Volume and Issue: 61, P. 1 - 13

Published: Jan. 1, 2023

Hyperspectral image (HSI) change detection is a technique for detecting the changes between multitemporal HSIs of same scene. Many existing methods have achieved good results, but there still exist problems as follows: 1) mixed pixels in HSI due to low spatial resolution hyperspectral sensor and other external interference 2) many deep learning-based networks cannot make full use correlation difference information bitemporal images. These are not conducive further improving accuracy detection. In this article, we propose an abundance matrix analysis network based on hierarchical multihead self-cross-hybrid attention (AMCAN-HMSchA) detection, which hierarchically highlights at subpixel level detect subtle changes. The endmember sharing-based learning module (AMLM) maps changed corresponding matrices. MSchA extracts enhanced features by constantly comparing self-correlation with cross matrices HSIs. Then, concatenated fed into fully connected layers obtain map. Experiments three widely used datasets show that proposed method has superior performance compared state-of-the-art methods.

Language: Английский

---

Efficient Inductive Vision Transformer for Oriented Object Detection in Remote Sensing Imagery

IEEE Transactions on Geoscience and Remote Sensing, Journal Year: 2023, Volume and Issue: 61, P. 1 - 20

Published: Jan. 1, 2023

Object detection is a fundamental task in remote sensing image analysis and scene understanding. Previous object detectors are typically based on convolutional neural networks (CNNs), whose performance significantly limited by the intrinsic locality of convolution operations. The emergence vision Transformers brings potential solutions to this problem, which have capability be solid alternative CNNs. However, three crucial obstacles hinder application detection, i.e., 1) high computational complexity, especially for high-resolution images, 2) training-and sample-inefficiency caused lack inductive bias, 3) difficulty learning arbitrary orientation knowledge geospatial objects. To address these issues, paper, novel efficient Transformer framework proposed oriented imagery. This follows hierarchical feature pyramid structure makes threefold contributions, as follows. Spatial redundancy images fully explored an adaptive multi-grained routing mechanism facilitate token sparsity Transformers, can dramatically reduce cost without comprising accuracy. A compact dual-path encoding architecture, where both global long-range dependencies local semantic relations jointly complementarily captured, enhance bias Transformers. An angle tokenization technique promote encoding, embedding, direction objects scenarios. In work, above contributions instantiated advanced Transformer-based detector, namely EIA-PVT. Comprehensive experiments two publicly available datasets demonstrated its effectiveness superiority images.

Language: Английский

---

RS³Mamba: Visual State Space Model for Remote Sensing Image Semantic Segmentation

IEEE Geoscience and Remote Sensing Letters, Journal Year: 2024, Volume and Issue: 21, P. 1 - 5

Published: Jan. 1, 2024

Language: Английский

---

Trustworthy remote sensing interpretation: Concepts, technologies, and applications

ISPRS Journal of Photogrammetry and Remote Sensing, Journal Year: 2024, Volume and Issue: 209, P. 150 - 172

Published: Feb. 8, 2024

Language: Английский

---

Enhancing UAV Aerial Image Analysis: Integrating Advanced SAHI Techniques With Real-Time Detection Models on the VisDrone Dataset

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 21621 - 21633

Published: Jan. 1, 2024

This research presents a groundbreaking approach in aerial image analysis by integrating the Real-Time Detection and Recognition (RT-DETR-X) model with Slicing Aided Hyper Inference (SAHI) methodology, utilizing VisDrone-DET dataset. Aimed at enhancing efficiency of drone technology across spectrum applications, including water conservancy, geological exploration, military operations, this study focuses on harnessing real-time, end-to-end object detection capabilities RT-DETR-X. Characterized its high-speed high-accuracy performance, particularly UAV photography, RT-DETR-X demonstrates remarkable 54.8% Average Precision (AP) 74 frames per second (FPS), surpassing similar models both speed accuracy. The thoroughly examines dataset, which encompasses diverse range small targets photography scenes. Covering 10 distinct categories, dataset provides robust platform for rigorous testing. emphasizes utilization original comprehensive training evaluation, alongside practical implementation SAHI method enhanced small-scale objects. Through an in-depth exploration model's performance various scenarios detailed environmental setup, paper underscores impact RT-DETR approach. findings reveal significant progress technologies, offering holistic framework effective efficient surveillance. integration not only boosts accuracy but also opens new avenues advanced applications.

Language: Английский

---