Jie Liu

Jie Liu

Auto-generated profile, based on publicly available information

Jie Liu

College of Artificial Intelligence and Automation, Hohai University, Changzhou, Jiangsu, China

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

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

Representative Publications

View All Publications
26 November 2025 in IEEE Transactions on Automation Science and Engineering
Verified
Journal Article
Closed access
YZYonghao Zhao Yonghao Zhao
Yonghao Zhao Yonghao Zhao
Jianjun NiJianjun Ni
Jianjun Ni
JLJie Liu
Jie Liu
YGYang Gu
Yang Gu
Simon X. YangSimon X. Yang
Simon X. Yang
Autonomous exploration in unknown environments is a critical capability for multi-UAV systems. However, existing methods often suffer from unbalanced task allocation, low exploration efficiency, and unstable paths, especially in large-scale and complex scenarios. To address these challenges, this paper presents an adaptive coordination exploration approach for multi-UAV systems. In the proposed approach, dynamic region allocation, entropy-guided local planning, and direction-consistent frontier selection are integrated to achieve efficient and collaborative exploration. The system first partitions the environment adaptively based on workload and regional complexity. It then prioritizes high-information-value areas for exploration. Directional constraints are further applied to improve path continuity and reduce turning redundancy. Extensive experiments in indoor maze and pillar environments, and outdoor forest and urban environments demonstrate that the proposed approach outperforms state-of-the-art baselines. Furthermore, ablation studies validate the necessity and complementarity of each module. This work provides a practical and efficient solution for multi-UAV exploration in structured and unstructured environments.

12 January 2026 in IEEE Sensors Journal
Verified
Journal Article
Closed access
JLJie Liu
Jie Liu
Jianjun NiJianjun Ni
Jianjun Ni
ZZZiru Zhang
Ziru Zhang
YGYang Gu
Yang Gu
Simon X. YangSimon X. Yang
Simon X. Yang
The infrared (IR)–visible (VIS) fusion-based object detection for uncrewed aerial vehicle (UAV) applications aims to achieve robust all-weather surveillance by integrating complementary information from IR and VIS modalities. However, the inherent heterogeneity and dynamic quality imbalance between modalities significantly hamper effective cross-modal feature utilization. In this paper, we propose a multimodal adaptive fusion-based Transformer network (MAFTNet) for effective multimodal feature learning and fusion. In the proposed model, a dual-branch modality synergy Transformer (DB-MST) module is designed, which strategically recombines modal features through hybrid modality channels and Transformer-based interaction, enhancing intermodal synergy. Then, a condition-driven channel reconstruction (CDCR) module is presented, which functions as a context-aware refiner, dynamically recalibrating channel contributions to highlight salient information while suppressing redundancies. Finally, a hybrid feature fusion encoder (HFFE) module is proposed, which performs hierarchical fusion and adaptive selection of multiscale, multilevel features to generate a unified representation. Experimental results demonstrate that our proposed MAFTNet achieves superior detection accuracy and inference speed compared with state-of-the-art (SOTA) methods.

15 May 2026 in IEEE Geoscience and Remote Sensing Letters
Verified
Journal Article
Closed access
Jianjun NiJianjun Ni
Jianjun Ni
JHJiabin Huang
Jiabin Huang
GTGuangyi Tang
Guangyi Tang
JLJie Liu
Jie Liu
WCWeidong Cao
Weidong Cao
PSPengfei Shi
Pengfei Shi
Object detection in remote sensing images has become a focal point in computer vision research. With the advancement of remote sensing technology, there is an increasing demand for detecting objects at arbitrary orientations. However, objects in remote sensing images often exhibit diverse scales and random rotations. Furthermore, rotated objects typically have high aspect ratios, making bounding box predictions sensitive to angular variations and increasing the complexity of detection. To address these challenges, this paper proposes an improved rotated object detection model based on geometric and size relationships. Firstly, an improved ResNet50 network based on multi-scale perception enhancement is proposed, where a feature pyramid network (FPN) and a channel attention mechanism are combined to enhance object features. Then, a coarse-to-fine optimization strategy is employed to precisely detect the rotated object. In the coarse stage, the rotated angle is computed by predicting the geometric relationship between the center points and boundary center points, which reduces the issues of angle periodicity and edge exchangeability. In the refined stage, an adaptive rotated intersection-over-union (IoU) loss is proposed, which incorporates the object’s aspect ratio and area, significantly improving the model’s adaptability across various scales, particularly excelling in high-aspect-ratio objects. Experimental results demonstrate that the proposed model effectively identifies rotated objects and outperforms existing models.

13 October 2026 in Measurement Science and Technology
Verified
Journal Article
Closed access
Jianjun NiJianjun Ni
Jianjun Ni
SWSheng Wu
Sheng Wu
JLJie Liu
Jie Liu
YGYang Gu
Yang Gu
PSPengfei Shi
Pengfei Shi
3D object detection is an essential and highly challenging task for unmanned autonomous systems operating in indoor scenes. Current mainstream 3D detection approaches rely on the direct encoding of point cloud coordinates, while neglecting the underlying geometric priors, thereby limiting the expressiveness of learned features. To deal with these problems, an efficient indoor 3D object detection model (Geo3D) is proposed in this paper, which exploits geometric priors to refine the feature learning pipeline. Specifically, during the initial stage of feature extraction, a geometric invariant embedding module is designed to compute geometric invariants that are robust to translation and rotation. These invariants provide structural priors of the point cloud distributions, and are embedded into voxel-wise feature representations. Subsequently, a hierarchical encoder is presented to capture multi-scale contextual features, augmented with geometric-aware refinement layers, where geometric information serves as conditional priors to guide fine-grained feature learning. Then, a synergistic attention calibration module is integrated into the generative decoder, which utilizes staged attention mechanisms to adaptively align high-layer semantic features with low-layer geometric details. Finally, extensive experiments are conducted on the SUN RGB-D and ScanNet V2 datasets, and the results demonstrate that the proposed Geo3D model achieves superior performance compared to existing point cloud–based detection methods.