SZShiliang Zhu

Shiliang Zhu

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Shiliang Zhu has a total of 10 co-authors. Here are the top co-authors by number of shared publications:

  • Zhensong Li

    Beijing Information Science and Technology University

    6 shared publications
  • Min Miao

    Academy of Smart IC and Network, Beijing Information Science and Technology University, Beijing, China

    4 shared publications
  • Bin Luo

    State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, China

    3 shared publications
  • Yutong Wang

    Center for Target Cognition Information Processing Science and Technology, College of Information and Communication Engineering, Beijing Information Science and Technology University, Beijing 100080, China

    3 shared publications
  • Yutong Wang

    Beijing Information Science and Technology University

    2 shared publications
  • Haoran Sun

    School of Information and Communication Engineering, The Center for Target Cognition Information Processing Science and Technology, Beijing Information Science and Technology University, Beijing 102206, China

    1 shared publication

Latest Publications

  • BSOEDet: Background Suppression and Object Enhancement Detector for UAV Aerial Imagery

    by Shiliang Zhu, Bin Luo, Jun Liu, Zhensong Li

    Uncrewed aerial vehicles (UAVs), leveraging their flexible low-altitude imaging and real-time data acquisition capabilities, are widely employed for aerial imagery tasks. However, UAV aerial images contain numerous small objects characterized by low pixel proportions and weak features, making them interference to background noise. To address this challenge, this paper proposes a UAV Aerial Imagery Background Suppression and Object Enhancement Detector (BSOEDet). Specifically, an Object and Background Aware Feature Extractor (OBAE) was developed, in which an auxiliary region classifier differentiates between background and object regions, and self-attention and channel attention mechanisms are combined to suppress irrelevant background and enhance object features, thereby maximizing the representation of small object feature. Furthermore, we propose a Deep Semantic Mining Small Object Module (DSMSOM). DSMSOM employs cross-attention to establish cross-spatial semantic links between deep and shallow features. By leveraging scale-invariant features between large and small objects, and semantic correlations among different objects, it mines potential small object features in the shallow layers using the semantic information of deep large objects, and further reinforces these representations with Direction Adaptive Convolution (DAConv). Moreover, a Global–Local Object Distribution Aware Classification Loss ( ${\mathcal {L}}_{\mathrm {GLOD}}$ ) is incorporated, in which the average object area and number of the entire train dataset and each batch are statistically computed to dynamically generate classification loss weights, thus enhancing the optimization contribution of small objects and few-shot classes during training. BSOEDet is validated on the VisDrone and UAVDT datasets. n the overall performance metric $\mathrm {mAP}_{50}$ , BSOEDet achieves improvements of 6.5% and 9.4% over the baseline models, and on the small object performance metric $\mathrm {mAP}_{s}$ , improvements of 4.6% and 8.4% are observed, demonstrating the effectiveness and robustness of these proposed approaches.


  • Open Access

    Region-aware and cross-scale feature mining for UAV object detection

    by Shiliang Zhu, Bin Luo, Zhensong Li

    1. In recent years, breakthrough progress in deep learning has driven significant development in the field of general object detection. Consequently, deep learning-based detection methods have been...

  • Open Access

    DPF-DETR: Enhancing Drone Image Detection with Density Perception and Multi-Scale Feature Fusion

    by Sidi Lai, Zhensong Li, Xiaotan Wei, Yutong Wang, Shiliang Zhu

    The DPF-DETR model has been designed to address the challenges encountered in object detection within drone imagery, particularly in scenarios involving significant target scale variations, dense targets, and complex backgrounds. To overcome the limitations of traditional object detection methods, the Density Sensing Mechanism (DSM) and Adaptive Density Map Loss (AdaptiveDM Loss) have been incorporated into the model to provide fine-grained supervision signals. The DSM optimizes the query selection mechanism by utilizing density maps, enabling the number of queries to be adaptively adjusted based on the distribution density of targets, thus improving detection accuracy in dense regions. Furthermore, the precision of the model in detecting dense targets is enhanced by AdaptiveDM Loss, which dynamically adjusts the weights for object localization and classification. Multi-scale feature fusion capabilities are also improved by the Multi-Scale Feature Fusion Network (MSFFN) and the Selective Feature Integration Module (SFIM). The MSFFN refines the fusion of features, which improves the detection of targets across various scales, particularly in complex scenes. Additionally, SFIM enhances the detection accuracy for small targets and complex backgrounds by integrating low-level spatial features with high-level semantic information. The Context-Sensitive Feature Interaction Module (CSFIM) further optimizes multi-scale feature fusion through context-guided interactions, bridging the semantic gap between features of different scales, thus improving the robustness of the model in dense scenarios. Experimental results have shown that DPF-DETR outperforms traditional models and state-of-the-art detection methods across multiple datasets, demonstrating superior robustness and accuracy, especially in dense target detection and complex background scenarios.


  • Open Access

    AutoMamba: Efficient Autonomous Driving Segmentation Model with Mamba

    by Haoran Sun, Zhensong Li, Shiliang Zhu

    Semantic segmentation for autonomous driving demands balancing high-fidelity perception with real-time latency. While Transformers achieve state-of-the-art results, their quadratic complexity bottlenecks high-resolution processing. State Space Models (SSMs) like Mamba offer linear complexity but often suffer from local detail loss and inefficient scanning strategies. We introduce AutoMamba, a tailored Hybrid-SSM architecture. We propose a Hybrid-SSM block incorporating Depthwise Convolutions to inject local spatial priors and a Stage-Adaptive Mixed-Scanning strategy. This strategy prioritizes horizontal context in early stages for road layouts while only activating vertical scanning in deep layers to preserve anisotropic structures like poles. Furthermore, we reveal that unlike Transformers, Mamba architectures require Auxiliary Supervision and Online Hard Example Mining (OHEM) to address “long-tail forgetting.” Experiments on Cityscapes and BDD100K under a training-from-scratch setting demonstrate AutoMamba’s superiority. Notably, AutoMamba-B0 achieves 67.79% mIoU on Cityscapes with 31.3% fewer FLOPs than SegFormer-B0. Moreover, while the larger SegFormer-B2 fails with Out-Of-Memory errors at 2048×2048 resolution, AutoMamba-B2 scales efficiently, validating its linear complexity advantage for next-generation perception systems.


  • TE-Net: remote sensing image detection network with resistance to background

    by Yutong Wang, Zhensong Li, Shiliang Zhu, Xiaotan Wei

  • Open Access

    L-ENet: An Ultralightweight SAR Image Detection Network

    by Yutong Wang, Min Miao, Shiliang Zhu

    Synthetic Aperture Radar (SAR) imagery plays a vital role in maritime vessel detection, despite facing challenges like complex marine environments, low-contrast target recognition, and similarities between ships and sea surface waves. Additionally, the deployment of traditional large models on radar-equipped edge devices is hampered by their extensive computational demands. Addressing these challenges, this study presents an ultra-lightweight network Lightweight Enhanced Network(L-ENet), improving upon YOLOv5n. The backbone network is switched to a more efficient ShufflenetV2, integrated with an improved attention mechanism, DCFAM (Dual-Core Fusion Attention Mechanism), for effective inter-channel information balancing. In the network's neck, the original C3 module has been replaced with the C3EGhost convolution. This convolution module integrates the Ghost and ECA (Efficient Channel Attention) attention mechanisms, aiming to mitigate potential accuracy loss during the lightweight process. Furthermore, this study introduces a novel multi-scale fusion pathway and a Concat module with adaptive weights to better harmonize semantic and detail information. Further, the tri-head detection structure is revised to a dual-head structure OMNI-DIMENSIONAL Adaptive Spatial Feature Fusion (ODASFF), using ODConv to allocate weights across different scales for efficient detection. Experimental results show that L-ENet has a computational cost of 0.6M and a parameter count of 2.1GFLOPs. These figures are 65% and 50% lower than those of the original YOLOv5n while maintaining the same detection accuracy of 97.8%. Hence, L-ENet offers an efficient and viable solution for maritime target detection, showcasing commendable detection performance in comparison with advanced models.


  • Open Access

    Lightweight high-precision SAR ship detection method based on YOLOv7-LDS

    by Shiliang Zhu, Min Miao

    The current challenges in Synthetic Aperture Radar (SAR) ship detection tasks revolve around handling significant variations in target sizes and managing high computational expenses, which hinder practical deployment on satellite or mobile airborne platforms. In response to these challenges, this research presents YOLOv7-LDS, a lightweight yet highly accurate SAR ship detection model built upon the YOLOv7 framework. In the core of YOLOv7-LDS’s architecture, we introduce a streamlined feature extraction network that strikes a delicate balance between detection precision and computational efficiency. This network is founded on Shufflenetv2 and incorporates Squeeze-and-Excitation (SE) attention mechanisms as its key elements. Additionally, in the Neck section, we introduce the Weighted Efficient Aggregation Network (DCW-ELAN), a fundamental feature extraction module that leverages Coordinate Attention (CA) and Depthwise Convolution (DWConv). This module efficiently aggregates features while preserving the ability to identify small-scale variations, ensuring top-quality feature extraction. Furthermore, we introduce a lightweight Spatial Pyramid Dilated Convolution Cross-Stage Partial Channel (LSPHDCCSPC) module. LSPHDCCSPC is a condensed version of the Spatial Pyramid Pooling Cross-Stage Partial Channel (SPPCSPC) module, incorporating Dilated Convolution (DConv) as a central component for extracting multi-scale information. The experimental results show that YOLOv7-LDS achieves a remarkable Mean Average Precision (mAP) of 99.1% and 95.8% on the SAR Ship Detection Dataset (SSDD) and the NWPU VHR-10 dataset with a parameter count (Params) of 3.4 million, a Giga Floating Point Operations Per Second (GFLOPs) of 6.1 and an Inference Time (IT) of 4.8 milliseconds. YOLOv7-LDS effectively strikes a fine balance between computational cost and detection performance, surpassing many of the current state-of-the-art object detection models. As a result, it offers a more resilient solution for maritime ship monitoring.


  • LRSNet: a high-efficiency lightweight model for object detection in remote sensing

    by Shiliang Zhu, Min Miao, Yutong Wang

    Unmanned aerial vehicles (UAVs) exhibit the ability to flexibly conduct aerial remote-sensing imaging. By employing deep learning object-detection algorithms, they efficiently perceive objects, finding widespread application in various practical engineering tasks. Consequently, UAV-based remote sensing object detection technology holds considerable research value. However, the background of UAV remote sensing images is often complex, with varying shooting angles and heights leading to difficulties in unifying target scales and features. Moreover, there is the challenge of numerous densely distributed small targets. In addition, UAVs face significant limitations in terms of hardware resources. Against this background, we propose a lightweight remote sensing object detection network (LRSNet) model based on YOLOv5s. In the backbone of LRSNet, the lightweight network MobileNetV3 is used to substantially reduce the model’s computational complexity and parameter count. In the model’s neck, a multiscale feature pyramid network named CM-FPN is introduced to enhance the detection capability of small objects. CM-FPN comprises two key components: C3EGhost, based on GhostNet and efficient channel attention modules, and the multiscale feature fusion channel attention mechanism (MFFC). C3EGhost, serving as CM-FPN’s primary feature extraction module, possesses lower computational complexity and fewer parameters, as well as effectively reducing background interference. MFFC, as the feature fusion node of CM-FPN, can adaptively weight the fusion of shallow and deep features, acquiring more effective details and semantic information for object detection. LRSNet, evaluated on the NWPU VHR-10, DOTA V1.0, and VisDrone-2019 datasets, achieved mean average precision of 94.0%, 71.9%, and 35.6%, with Giga floating-point operations per second and Param (M) measuring only 5.8 and 4.1, respectively. This outcome affirms the efficiency of LRSNet in UAV-based remote-sensing object detection tasks.


  • SCNet: A Lightweight and Efficient Object Detection Network for Remote Sensing

    by Shiliang Zhu, Min Miao

    Detecting small objects in remote sensing images is meaningful challenging, especially when deploying existing object detection models on edge terminal devices with limited hardware resources. In this study, we present an efficient remote sensing object detection model named SCNet, based on the ultra-lightweight YOLOv5n. To address the significant feature loss issue in small objects within the model’s neck, we introduce the Selective Feature Enhancement Block (SFEB). The SFEB selectively processes a portion of feature maps that contribute more to semantic information extraction while retaining another portion, enabling us to extract rich semantic information while preserving crucial details information necessary for small object detection. Furthermore, we incorporate the Contextual Transformer Block (CTB) at the neck and Backbone junction, which enhances the model’s ability to understand relationships and boundaries between object s and backgrounds by exploring contextual information in shallow-level feature maps. This improves the model’s capability to detect challenging small and medium objects. Experimental results on the NWPU VHR-10 and DIOR datasets demonstrate the model’s performance, achieving mean Average Precisions (mAP) of 96.6% and 72.6% at IOU=0.5. The model operates at 487 Frames Per Second with a batch size of 32 (FPS32), requiring only 4.6 Giga Floating-point Operations Per second (GFLOPs) and 1.8 million params.