A5015745132- Advanced Neural Network Applications
- Video Surveillance and Tracking Methods
- Advanced Image and Video Retrieval Techniques
- Human Pose and Action Recognition
- Face and Expression Recognition
- Domain Adaptation and Few-Shot Learning
- Imbalanced Data Classification Techniques
- Remote-Sensing Image Classification
- Industrial Vision Systems and Defect Detection
- Machine Learning and Data Classification
- Face recognition and analysis
- Remote Sensing and Land Use
- Privacy-Preserving Technologies in Data
- Cellular and Composite Structures
- Text and Document Classification Technologies
- Anomaly Detection Techniques and Applications
- Advanced Vision and Imaging
- Biomimetic flight and propulsion mechanisms
- Advanced Graph Neural Networks
- Robotic Locomotion and Control
- Image Enhancement Techniques
- Machine Learning and ELM
- Multimodal Machine Learning Applications
- Human Mobility and Location-Based Analysis
- Topic Modeling
Northwestern Polytechnical University
2018-2025
Xiamen University
2019-2025
Xiamen University of Technology
2022-2025
Jinan University
2025
Fuzhou University
2024
Tianjin University of Technology
2024
Kaili University
2024
Hong Kong Baptist University
2015-2024
China Electric Power Research Institute
2024
Jimei University
2022-2024
Visible-infrared person re-identification (VI-ReID) aims to search identities of pedestrians across different spectra. In this task, one the major challenges is modality discrepancy between visible (VIS) and infrared (IR) images. Some state-of-the-art methods try design complex networks or generative mitigate while ignoring highly non-linear relationship two modalities VIS IR. paper, we propose a middle generator (MMG), which helps reduce discrepancy. Our MMG can effectively project IR...
Recent research has shown that utilizing the spectral-spatial information can improve performance of hyperspectral image (HSI) classification. Since HSI is a 3-D cube datum, spatial filtering becomes simple and effective method for extracting information. In this paper, we propose scattering wavelet transform, which filters data with cascade decompositions, complex modulus, local weighted averaging. The feature adequately capture classification step, support vector machine based on Gaussian...
To efficiently improve the accuracy of hyperspectral image (HSI) classification, spatial information is usually fused with spectral so that classification performance can be enhanced. In this paper, we propose a new method called wavelet transform-based smooth ordering (WTSO). WTSO consists three main components: transform for feature extraction, spectral–spatial based similarity measurement, 1D embedding, and construction final classifier using interpolation scheme. Specifically, first...
Long-tailed data is still a big challenge for deep neural networks, even though they have achieved great success on balanced data. We observe that vanilla training longtailed with crossentropy loss makes the instance-rich head classes severely squeeze spatial distribution of tail classes, which leads to difficulty in classifying class samples. Furthermore, original can only propagate gradient short-lively because softmax form rapidly approaches zero as logit difference increases. This...
Deep neural networks have made huge progress in the last few decades. However, as real-world data often exhibits a long-tailed distribution, vanilla deep models tend to be heavily biased toward majority classes. To address this problem, state-of-the-art methods usually adopt mixture of experts (MoE) focus on different parts distribution. Experts these are with same model depth, which neglects fact that classes may preferences fit by depths. end, we propose novel MoE-based method called...
The imbalanced distribution of long-tailed data presents a considerable challenge for deep learning models, as it causes them to prioritize the accurate classification head classes but largely disregard tail classes. biased decision boundary caused by inadequate semantic information in is one key factors contributing their low recognition accuracy. To rectify this issue, we propose augment grafting diverse from classes, referred head-to-tail fusion (H2T). We replace portion feature maps with...
This paper presents two origin-destination flow estimation models using sampled GPS positions of probe vehicles and link counts. The first model, named as SPP model (scaled OD prior OD), uses scaled vehicle matrix applies conventional generalized least squares (GLS) framework to conduct correction counts; the second PRA (probe ratio assignment), is an extension in which observed ratios are also included additional information process. For both models, study explored a new way construct...
Recent studies of imbalanced data classification have shown that the imbalance ratio (IR) is not only cause performance loss in a classifier, as other factors, such small disjuncts, noise, and overlapping, can also make problem difficult. The relationship between IR factors has been demonstrated, but to best our knowledge, there no measurement extent which class influences data. In addition, it unknown factor serves main barrier for set. this article, we focus on Bayes optimal classifier...
One of the most challenging problems in field online learning is concept drift, which deeply influences classification stability streaming data. If data stream imbalanced, it even more difficult to detect drifts and make an learner adapt them. Ensemble algorithms have been found effective for with whereby individual classifier built each incoming chunk its associated weight adjusted manage drift. However, adjust weights achieve a balance between adaptability ensemble classifiers. In...
Computer vision has achieved impressive progress in recent years. Meanwhile, mobile phones have become the primary computing platforms for millions of people. In addition to phones, many autonomous systems rely on visual data making decisions, and some these limited energy (such as unmanned aerial vehicles also called drones robots). These batteries, efficiency is critical. This paper serves following two main purposes. First, examine state art low-power solutions detect objects images....
Class imbalance problem has been extensively studied in the recent years, but imbalanced data clustering unsupervised environment, that is, number of samples among clusters is imbalanced, yet to be well studied. This paper, therefore, studies within framework k-means-type competitive learning. We introduce a new method called self-adaptive multiprototype-based learning (SMCL) for clusters. It uses multiple subclusters represent each cluster with an automatic adjustment subclusters. Then, are...
In this work, we construct a large-scale dataset for vehicle re-identification (ReID), which contains 137k images of 13k instances captured by UAV-mounted cameras. To our knowledge, it is the largest UAV-based ReID dataset. increase intra-class variation, each at least two UAVs different locations, with diverse view-angles and flight-altitudes. We manually label variety attributes, including type, color, skylight, bumper, spare tire luggage rack. Furthermore, image, annotator also required...
It is not uncommon that real-world data are distributed with a long tail. For such data, the learning of deep neural networks becomes challenging because it hard to classify tail classes correctly. In literature, several existing methods have addressed this problem by reducing classifier bias, provided features obtained long-tailed representative enough. However, we find training directly on leads uneven embedding space. That is, space head severely compresses classes, which conducive...
Concept drifts occurring in data streams will jeopardize the accuracy and stability of online learning process. If stream is imbalanced, it be even more challenging to detect cure concept drift. In literature, these two problems have been intensively addressed separately, but yet well studied when they occur together. this paper, we propose a chunk-based incremental method called Dynamic Weighted Majority for Imbalance Learning (DWMIL) deal with drift class imbalance problem. DWMIL utilizes...
Pedestrian attribute recognition (PAR) has received increasing attention because of its wide application in video surveillance and pedestrian analysis. Extracting robust feature representation is one the key challenges this task. The existing methods primarily rely on convolutional neural networks (CNNs) as backbone network for extraction. However, these mainly focus small discriminative regions while ignoring global perspective. To overcome limitations, we propose PARFormer, a pure...
The previously known works studying the generalization ability of support vector machine classification (SVMC) algorithm are usually based on assumption independent and identically distributed samples. In this paper, we go far beyond classical framework by SVMC uniformly ergodic Markov chain (u.e.M.c.) We analyze excess misclassification error u.e.M.c. samples, obtain optimal learning rate for also introduce a new sampling to generate samples from given dataset, present numerical studies...
Purpose This paper aims to imitate a cownose ray develop fish robot with paired flexible multi-fin-ray oscillating pectoral fins (OPFs) and control it accomplish vivid stable 3-D motions using central pattern generators (CPGs) fuzzy algorithm. Design/methodology/approach The ray’s asymmetric sine-like oscillations were analyzed. Then cownose-ray-like named Robo-ray was developed, which has OPFs actively the fin shape two tail depth. To solve problem of coordinated for multi-degree-of-freedom...
Federated learning provides a privacy guarantee for generating good deep models on distributed clients with different kinds of data. Nevertheless, dealing non-IID data is one the most challenging problems federated learning. Researchers have proposed variety methods to eliminate negative influence non-IIDness. However, they only focus provided that universal class distribution balanced. In many real-world applications, long-tailed, which causes model seriously biased. Therefore, this paper...
Sparsity-based models have been widely applied to hyperspectral image (HSI) classification. The class label of the test sample is determined by minimum residual error based on sparse vector, which viewed as a pattern original in sparsity-based model. From aspect classification, similar samples same should patterns. However, due independent reconstruction process, similarity among vectors these lost. To enforce such information, regularized representation (RSR) model proposed. First,...
Label noise is one of the key factors that lead to poor generalization deep learning models. Existing label-noise methods usually assume ground-truth classes training data are balanced. However, real-world often imbalanced, leading inconsistency between observed and intrinsic class distribution with label noises. In this case, it hard distinguish clean samples from noisy on tail unknown distribution. paper, we propose a frame-work for intrinsically long-tailed data. Specifically, two-stage...