Shenglu Zhao

Shenglu Zhao

Auto-generated profile, based on publicly available information

Shenglu Zhao

Shandong University of Science and Technology, College of Computer Science and Engineering, Qingdao, China

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

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

Representative Publications

View All Publications
12 October 2023 in Expert Systems with Applications
Verified
Journal Article
Closed access
XZXingLi Zhang
XingLi Zhang
LLLei Liang
Lei Liang
SZShenglu Zhao
Shenglu Zhao
ZWZhihui Wang
Zhihui Wang

1 May 2025 in Journal of Electronic Imaging
Verified
Journal Article
Closed access
PDF file
XZXingli Zhang
Xingli Zhang
SZShenglu Zhao
Shenglu Zhao
LLLei Liang
Lei Liang
SYShansong Yang
Shansong Yang
ZWZhihui Wang
Zhihui Wang
Visual segmentation has emerged as a prominent research frontier, with widespread applications driving technological advancements in traditional industries and everyday life. We focus on the segmentation of tactile paving, a critical infrastructure element designed to assist visually impaired individuals in navigation. To enhance the accuracy and robustness of tactile paving segmentation and overcome the limitations of traditional U-Net in complex scenarios and multi-scale feature processing, a basic reinforcement module, termed the convolutional-efficient attention module, is introduced. The module integrates convolutional layers, efficient channel attention (ECA), and atrous spatial pyramid pooling (ASPP) based on U-Net. The ECA is employed to enhance the focus of U-Net on the critical features of tactile paving, whereas ASPP further improves the capability of U-Net to extract multi-scale tactile paving features, thereby achieving superior feature extraction. The proposed approach begins with traditional convolutional layers to extract preliminary feature maps. The ECA module is then utilized to compute channel-wise importance and adaptively distribute weights, thereby capturing inter-channel dependencies and improving the network’s focus on significant features. Simultaneously, multi-scale feature extraction is performed using ASPP to expand the receptive field and incorporate contextual information, resulting in more precise feature maps. Furthermore, the loss function is optimized by introducing a Laplacian loss term, which measures the discrepancies between ground truth and predicted results. This optimization enhances the network’s capacity to capture subtle color and texture variations, leading to improved edge detection for tactile paving. Experimental evaluations demonstrate that the proposed method outperforms other advanced networks, achieving a mean intersection over union of 95.62% and an accuracy of 99.01%. These results establish the proposed method as a valuable contribution to tactile paving segmentation and provide a robust reference for future research in this domain.