Author name cluster

Xinying Wang

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

5 papers

1 author row

EAAI Journal 2026 Journal Article

Semantic-geometric dual knowledge guided instance segmentation for vehicle components

Zhenqi Zhang
Xunqi Zhou
Yongjie Zhai
Nianhao Chen
Qianming Wang
Xinying Wang

The detection and segmentation of vehicle components are crucial steps in intelligent vehicle damage assessment. However, due to the wide variety of vehicle components with diverse shapes and the high similarity between mirror-symmetric components, missed detections and false positives remain common challenges in vehicle component detection and segmentation. To address these challenges, this paper proposes a deep learning-based dual-knowledge guided vehicle component instance segmentation network. The proposed method fuses implicit multi-scale semantic knowledge by the enhanced semantic knowledge network, thereby improving the model’s capability to filter and capture critical component information. This design enhances focus on key foreground features while suppressing interference from background features. Furthermore, the proposed embedded geometric knowledge module synergizes geometric constraint knowledge of vehicle components with the model’s intrinsic learning capability. By reasoning about positions among three types of landmark components, it explicitly supplements the model with spatial relationship information of components that are inherently challenging to learn from data alone. We evaluate the detection and segmentation performance of the proposed method on a dataset comprising 59 classes of vehicle components. Compared with the baseline model, our method achieves significant improvements of 19. 3% and 19. 1% in detection mean average precision and segmentation mean average precision, respectively. Additionally, we evaluate the generalization capability of the proposed method on an independent public dataset. Compared with state-of-the-art instance segmentation methods, the proposed method demonstrates superior performance in both detection and segmentation tasks for vehicle components.

Details DOI

AAAI Conference 2025 Conference Paper

Two-stream Beats One-stream: Asymmetric Siamese Network for Efficient Visual Tracking

Jiawen Zhu
Huayi Tang
Xin Chen
Xinying Wang
Dong Wang
Huchuan Lu

Efficient tracking has garnered attention for its ability to operate on resource-constrained platforms for real-world deployment beyond desktop GPUs. Current efficient trackers mainly follow precision-oriented trackers, adopting a one-stream framework with lightweight modules. However, blindly adhering to the one-stream paradigm may not be optimal, as incorporating template computation in every frame leads to redundancy, and pervasive semantic interaction between template and search region places stress on edge devices. In this work, we propose a novel asymmetric Siamese tracker named AsymTrack for efficient tracking. AsymTrack disentangles template and search streams into separate branches, with template computing only once during initialization to generate modulation signals. Building on this architecture, we devise an efficient template modulation mechanism to unidirectional inject crucial cues into the search features, and design an object perception enhancement module that integrates abstract semantics and local details to overcome the limited representation in lightweight tracker. Extensive experiments demonstrate that AsymTrack offers superior speed-precision trade-offs across different platforms compared to the current state-of-the-arts. For instance, AsymTrack-T achieves 60.8% AUC on LaSOT and 224/81/84 FPS on GPU/CPU/AGX, surpassing HiT-Tiny by 6.0% AUC with higher speeds.

PDF Details DOI

EAAI Journal 2024 Journal Article

SU-VPDN: A scene understanding method for vehicle part detection

Yongjie Zhai
Nianhao Chen
Zhenqi Zhang
Xunqi Zhou
Xuan Liu
Xinying Wang
Qianming Wang

High-accuracy positioning and identification of vehicle parts are crucial for status detection, fault diagnosis, and the development of intelligent damage detection systems. However, detecting left–right and front-rear categories in vehicle parts remains challenging due to numerous categories and feature similarity. To address this, we propose the scene understanding vehicle part detection network (SU-VPDN). It comprises three subnetworks: vehicle orientation scene classification (VOSC), vehicle part instance segmentation (VPIS), and scene understanding (SU). VOSC performs preliminary classification of vehicle orientation scenes, while VPIS detects vehicle parts. SU fine-tunes VOSC results using the part logic correction module, fuses VOSC and VPIS results using the part replacement module, and obtains accurate VPIS results through confidence regulation and hierarchical filtering. SU-VPDN integrates results from multiple subnetworks, transforming the challenge of detecting similar objects into orientation scene detection for precise recognition. We evaluate the proposed method on a dataset containing 59 types of vehicle parts and compare it with advanced segmentation detection models. Experimental results demonstrate that SU-VPDN outperforms other models, achieving a 12. 2% and 12. 3% improvement in mean average precision (mAP) for box and mask, respectively, compared to the baseline model. Moreover, qualitative and quantitative experiments show that SU-VPDN significantly enhances the detection performance of objects with similar features. Overall, our proposed SU-VPDN offers a promising solution for detecting left–right and front-rear categories in vehicle parts. It surpasses existing models and can enable insurance companies to replace damaged parts and improve claims efficiency efficiently.

Details DOI

AAAI Conference 2020 Conference Paper

HDK: Toward High-Performance Deep-Learning-Based Kirchhoff Analysis

Xinying Wang
Olamide Timothy Tawose
Feng Yan
Dongfang Zhao

The Kirchhoff law is one of the most widely used physical laws in many engineering principles, e. g. , biomedical engineering, electrical engineering, and computer engineering. One challenge of applying the Kirchhoff law to real-world applications at scale lies in the high, if not prohibitive, computational cost to solve a large number of nonlinear equations. Despite recent advances in leveraging a convolutional neural network (CNN) to estimate the solutions of Kirchhoff equations, the low performance is still signiﬁcantly hindering the broad adoption of CNN-based approaches. This paper proposes a high-performance deep-learning-based approach for Kirchhoff analysis, namely HDK. HDK employs two techniques to improve the performance: (i) early pruning of unqualiﬁed input candidates and (ii) parallelization of forward labelling. To retain high accuracy, HDK also applies various optimizations to the data such as randomized augmentation and dimension reduction. Collectively, the aforementioned techniques improve the analysis speed by 8× with accuracy as high as 99. 6%.

PDF Details

EAAI Journal 2015 Journal Article

Improved extreme learning machine for multivariate time series online sequential prediction

Xinying Wang
Min Han

Multivariate time series has attracted increasing attention due to its rich dynamic information of the underlying systems. This paper presents an improved extreme learning machine for online sequential prediction of multivariate time series. The multivariate time series is first phase-space reconstructed to form the input and output samples. Extreme learning machine, which has simple structure and good performance, is used as prediction model. On the basis of the specific network function of extreme learning machine, an improved Levenberg–Marquardt algorithm, in which Hessian matrix and gradient vector are calculated iteratively, is developed to implement online sequential prediction. Finally, simulation results of artificial and real-world multivariate time series are provided to substantiate the effectiveness of the proposed method.

Details DOI