Tianyang Hu

Ms. Tianyang Hu

orcid

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
18 December 2023 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Verified
Journal Article
Open access
BMBoyi Ma
Boyi Ma
FWFalin Wu
Falin Wu
THTianyang Hu
Tianyang Hu
Loghman FathollahiLoghman Fathollahi
Loghman Fathollahi
XSXiaohong Sui
Xiaohong Sui
Byambakhuu GantumurByambakhuu Gantumur
Byambakhuu Gantumur
Multi-label classification in remote sensing is very significant which plays an important role in extracting valuable information from satellite imagery. Ignoring the distinct information provided by labels in each image or transforming images into content-aware category representations without considering inherent correlation of labels within the dataset can result in the establishment of improper relationships between images and labels, ultimately leading to a significant degradation in accuracy. To address this problem, this paper proposes a Label-Driven Graph Convolutional Network (LD-GCN) to excavate substantial information using the inherent correlation of labels from datasets and build a strong relationship between labels and images. The framework consists of two modules, i.e. label recognition GCN (LRGCN) and semantic enrichment module (SEM). The LRGCN module yields rich and valuable information from the inherent correlation of labels and builds a strong relationship between images and labels. The SEM further enriches the semantics obtained from LRGCN. Experiments conducted on UCM, AID, and DFC15 multi-label remote sensing datasets illustrate that LD-GCN outperforms the state-of-the-art methods on key evaluation metrics.

15 May 2024 in Remote Sensing
Verified
Journal Article
Open access
PDF file
FWFalin Wu
Falin Wu
THTianyang Hu
Tianyang Hu
YXYu Xia
Yu Xia
BMBoyi Ma
Boyi Ma
SSSaddam Sarwar
Saddam Sarwar
CZChunxiao Zhang
Chunxiao Zhang
Ships are important targets for modern naval warfare detection and reconnaissance. The accurate detection of ships contributes to the maintenance of maritime rights and interests and the realisation of naval strategy. Synthetic Aperture Radar (SAR) image detection tasks play a vital role in ship detection, which has consistently been a research hotspot in the field of SAR processing. Although significant progress has been achieved in SAR ship detection techniques using deep learning methods, some challenges still persist. Natural images and SAR images significantly diverge in imaging mechanisms and scattering characteristics. In complex background environments, ships exhibit multiscale variations and dense arrangements, and numerous small-sized ships may be present, culminating in false or missed detections. To address these issues, we propose a novel SAR ship detection network, namely, a Wavelet-Driven Feature-Enhanced Attention–You Only Look Once X (WDFA-YOLOX) network. Firstly, we propose a Wavelet Cascade Residual (WCR) module based on the traditional image processing technique wavelet transform, which is embedded within an improved Spatial Pyramid Pooling (SPP) module, culminating in the formation of the effective wavelet transform-based SPP module (WSPP). The WSPP compensates for the loss of fine-grained feature information during pooling, enhancing the capability of the network to detect ships amidst complex background interference. Secondly, a Global and Local Feature Attention Enhancement (GLFAE) module is proposed, leveraging a parallel structure that combines convolutional modules with transformer modules to reduce the effect of irrelevant information and effectively strengthens valid features associated with small-sized ships, resulting in a reduction in false negatives in small-sized ship detection. Finally, a novel loss function, the Chebyshev distance-generalised IoU loss function, is proposed to significantly enhance both the precision of the detection box and the network convergence speed. To support our approach, we performed thorough experiments on the SSDD and HRSID, achieving an average precision (AP) of 99.11% and 96.20%, respectively, in ship detection. The experimental results demonstrate that WDFA-YOLOX has significant advantages in terms of detection accuracy, generalisation capability, and detection speed and can effectively realise more accurate detection in SAR images, consistently exhibiting superior performance and application value in SAR ship detection.

2 December 2024 in IEEE Transactions on Affective Computing
Verified
Journal Article
Closed access
FWFalin Wu
Falin Wu
YXYu Xia
Yu Xia
BMBoyi Ma
Boyi Ma
THTianyang Hu
Tianyang Hu
JYJingyao Yang
Jingyao Yang
DHDi Huang
Di Huang
Micro-expressions (MEs) are spontaneous facial movements that reveal an individual's genuine emotions and play a crucial role in various domains, including lie detection, criminal analysis, mental health treatment, national security, and others. Micro-expression recognition is a highly complex aspect within the domain of affective computing, aimed at identifying subtle facial motions that are difficult for humans to discern accurately. To model the subtle facial muscle motions and the brief duration of MEs, we propose a robust micro-expression recognition (MER) network, named the attention mechanismbased motion magnification guided micro-expression recognition network (AM-MM-MER). This network consists of two primary components: the ST-MEMM network, which enhances subtle motions in micro-expression videos to reveal imperceptible facial muscle motions, and the AM-MER, which focuses on facial landmarks related to micro-expressions and incorporates novel landmark positions to extract the underlying relationships among these landmarks, thereby reducing interference from video magnification and irrelevant identity features. Extensive analysis on the CASME II and SAMM datasets demonstrates the high accuracy and effectiveness of the proposed network, achieving superior results compared to state-of-the-art methods. Ablation studies further illustrate the robustness of the proposed network.

31 January 2025 in Remote Sensing
Verified
Journal Article
Open access
PDF file
SSSaddam Sarwar
Saddam Sarwar
UKUsman Ahmed Khan
Usman Ahmed Khan
FWFalin Wu
Falin Wu
SHSarah Hasan
Sarah Hasan
MZMuhammad Zohaib
Muhammad Zohaib
THTianyang Hu
Tianyang Hu
In the past two decades, Islamabad has experienced significant urbanization. As a result of inadequate urban planning and spatial distribution, it has significantly influenced land use–land cover (LULC) changes and green areas. To assess these changes, there is an increasing need for reliable and appropriate information about urbanization. Landsat imagery is categorized into four thematic classes using a supervised classification method called the support vector machine (SVM): built-up, bareland, vegetation, and water. The results of the change detection of post-classification show that the city region increased from 6.37% (58.09 km2) in 2000 to 28.18% (256.49 km2) in 2020, while vegetation decreased from 46.97% (428.28 km2) to 34.77% (316.53 km2) and bareland decreased from 45.45% (414.37 km2) to 35.87% (326.49 km2). Utilizing a land change modeler (LCM), forecasts of the future conditions in 2025, 2030, and 2035 are predicted. The artificial neural network (ANN) model embedded in IDRISI software 18.0v based on a well-defined backpropagation (BP) algorithm was used to simulate future urban sprawl considering the historical pattern for 2015–2020. Selected landscape morphological measures were used to quantify and analyze changes in spatial structure patterns. According to the data, the urban area grew at a pace of 4.84% between 2015 and 2020 and will grow at a rate of 1.47% between 2020 and 2035. This growth in the metropolitan area will encroach further into vegetation and bareland. If the existing patterns of change persist over the next ten years, a drop in the mean Euclidian Nearest Neighbor Distance (ENN) of vegetation patches is anticipated (from 104.57 m to 101.46 m over 2020–2035), indicating an accelerated transformation of the landscape. Future urban prediction modeling revealed that there would be a huge increase of 49% in urban areas until the year 2035 compared to the year 2000. The results show that in rapidly urbanizing areas, there is an urgent need to enhance land use laws and policies to ensure the sustainability of the ecosystem, urban development, and the preservation of natural resources.

30 January 2023 in arXiv
Verified
Preprint
Open access
PDF file
LYLonghui Yu
Longhui Yu
THTianyang Hu
Tianyang Hu
LHLanqing Hong
Lanqing Hong
ZLZhen Liu
Zhen Liu
AWAdrian Weller
Adrian Weller
WLWeiyang Liu
Weiyang Liu
It has been observed that neural networks perform poorly when the data or tasks are presented sequentially. Unlike humans, neural networks suffer greatly from catastrophic forgetting, making it impossible to perform life-long learning. To address this issue, memory-based continual learning has been actively studied and stands out as one of the best-performing methods. We examine memory-based continual learning and identify that large variation in the representation space is crucial for avoiding catastrophic forgetting. Motivated by this, we propose to diversify representations by using two types of perturbations: model-agnostic variation (i.e., the variation is generated without the knowledge of the learned neural network) and model-based variation (i.e., the variation is conditioned on the learned neural network). We demonstrate that enlarging representational variation serves as a general principle to improve continual learning. Finally, we perform empirical studies which demonstrate that our method, as a simple plug-and-play component, can consistently improve a number of memory-based continual learning methods by a large margin.

17 October 2023 in arXiv
Verified
Preprint
Open access
PDF file
JMJiajun Ma
Jiajun Ma
THTianyang Hu
Tianyang Hu
WWWenjia Wang
Wenjia Wang
JSJiacheng Sun
Jiacheng Sun
Guidance in conditional diffusion generation is of great importance for sample quality and controllability. However, existing guidance schemes are to be desired. On one hand, mainstream methods such as classifier guidance and classifier-free guidance both require extra training with labeled data, which is time-consuming and unable to adapt to new conditions. On the other hand, training-free methods such as universal guidance, though more flexible, have yet to demonstrate comparable performance. In this work, through a comprehensive investigation into the design space, we show that it is possible to achieve significant performance improvements over existing guidance schemes by leveraging off-the-shelf classifiers in a training-free fashion, enjoying the best of both worlds. Employing calibration as a general guideline, we propose several pre-conditioning techniques to better exploit pretrained off-the-shelf classifiers for guiding diffusion generation. Extensive experiments on ImageNet validate our proposed method, showing that state-of-the-art diffusion models (DDPM, EDM, DiT) can be further improved (up to 20%) using off-the-shelf classifiers with barely any extra computational cost. With the proliferation of publicly available pretrained classifiers, our proposed approach has great potential and can be readily scaled up to text-to-image generation tasks. The code is available at https://github.com/AlexMaOLS/EluCD/tree/main.