Bushi Liu

Bushi Liu


Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

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

Representative Publications

View All Publications
12 June 2024 in Symmetry
Verified
Journal Article
Open access
PDF file
BLBushi Liu
Bushi Liu
CYCuiying Yu
Cuiying Yu
Bolun ChenBolun Chen
Bolun Chen
YZYue Zhao
Yue Zhao
In recent years, frequent chemical production safety incidents in China have been primarily attributed to dangerous behaviors by workers. Current monitoring methods predominantly rely on manual supervision, which is not only inefficient but also prone to errors in complex environments and with varying target scales, leading to missed or incorrect detections. To address this issue, we propose a deep learning-based object detection model, YOLO-GP. First, we utilize a grouped pointwise convolutional (GPConv) module of symmetric structure to facilitate information exchange and feature fusion in the channel dimension, thereby extracting more accurate feature representations. Building upon the YOLOv8n model, we integrate the symmetric structure convolutional GPConv module and design the dual-branch aggregation module (DAM) and Efficient Spatial Pyramid Pooling (ESPP) module to enhance the richness of gradient flow information and the capture of multi-scale features, respectively. Finally, we develop a channel feature enhancement network (CFE-Net) to strengthen inter-channel interactions, improving the model’s performance in complex scenarios. Experimental results demonstrate that YOLO-GP achieves a 1.56% and 11.46% improvement in the mAP@.5:.95 metric on a custom dangerous behavior dataset and a public Construction Site Safety Image Dataset, respectively, compared to the baseline model. This highlights its superiority in dangerous behavior object detection tasks. Furthermore, the enhancement in model performance provides an effective solution for improving accuracy and robustness, promising significant practical applications.

1 March 2026 in Engineering Research Express
Verified
Journal Article
Closed access
KCKailu Chang
Kailu Chang
BLBushi Liu
Bushi Liu
HZHaonan Zhang
Haonan Zhang
Bolun ChenBolun Chen
Bolun Chen
JZJianyang Zhao
Jianyang Zhao
Ship detection is vital for maritime safety and surveillance. However, traditional methods face challenges such as high false detection rates and difficulty in detecting small objects in complex backgrounds. To address these issues, we propose FAD-YOLO, an innovative model designed for ship object detection. FAD-YOLO adopts a Multi-scale Cross-layer Fusion Strategy that effectively integrates features across different scales, ensuring the capture of rich contextual information and enhancing ship recognition in complex scenarios. Additionally, a Dilated Convolution-enhanced Spatial Pyramid Pooling Fast module is introduced to extract hierarchical feature information, enhancing the accuracy of small object detection. The Enhanced Information Fusion Module further optimizes computation while preserving strong feature extraction capabilities. Compared to YOLOv8-N/S, FAD-YOLO-N/S improves mAP(50:95) by up to 5.8%, demonstrating superior performance in ship detection tasks. FAD-YOLO not only advances ship detection technology but also offers new insights into object detection methodologies in computer vision.

30 July 2026 in Measurement Science and Technology
Verified
Journal Article
Closed access
Bolun ChenBolun Chen
Bolun Chen
KCKailu Chang
Kailu Chang
YCYanming Chen
Yanming Chen
BLBushi Liu
Bushi Liu
JZJianyang Zhao
Jianyang Zhao
High-resolution remote sensing ship detection plays an indispensable role in maritime surveillance and intelligent ocean management; however, existing approaches still suffer from several fundamental technical bottlenecks. Large-kernel convolution-based methods struggle to simultaneously preserve fine-grained local feature representation and achieve effective multi-scale adaptability. Multi-branch enhancement strategies often incur inefficient feature collaboration and redundant computation, while lightweight detection models face inherent limitations in realizing deep feature complementarity and semantic enhancement under strict computational constraints. To address these challenges, this paper proposes a lightweight detection model, termed GLDet (Global-Local Receptive Field Enhanced Detector for Ship Detection), which integrates global perception and local feature enhancement. Specifically, the proposed GLDet incorporates an LKAttConv module, which combines large-kernel convolution with a spatial attention mechanism to strengthen local feature extraction while improving multi-scale adaptability. In addition, a C3K2LS module is designed with a streamlined multi-branch structure to enhance feature-path collaboration efficiency and effectively reduce computational redundancy. Furthermore, GLDet introduces a Poly kernel inception spatial pyramid pooling module to optimize multi-scale pooling and attention fusion, enabling deep semantic enhancement under lightweight constraints.Extensive experimental results demonstrate that GLDet achieves an mAP@0.5 of 98.5% and an mAP@0.5–0.95 of 85.4% on the HRSC dataset, with an inference speed of 134.51 FPS, significantly outperforming mainstream methods in terms of accuracy, efficiency, and lightweight balance. These results indicate that GLDet provides an effective and efficient solution for ship detection in complex remote sensing scenarios and offers valuable insights for the design of detection models targeting small-scale and morphologically diverse objects.