Baishun Su

Dr. Baishun Su


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
26 September 2025 in IEEE Transactions on Instrumentation and Measurement
Verified
Journal Article
Closed access
BSBaishun Su
Baishun Su
HLHao Li
Hao Li
To enable rapid automatic reading of various pointer-type meters in different environments, this study designs a universal multitask cascaded network, divided into three stages: meter detection, feature extraction, and reading. First, pointer-type meters are detected using YOLOv12. Then, the text features of pointer-type meters are extracted using the designed SFTR, while the proposed MSFFNet extracts pointer and scale line features. The Multi-directional Strip convolution(MDSConv) module in MSFFNet can extract strip features in different directions, and the Multi-scale fusion(MSF) can efficiently perform feature fusion. Finally, the reading is calculated using the designed Difference calibration (Dc) reading method, which, through a segmented angle compensation mechanism combined with a non-linear interpolation strategy, can adaptively correct errors caused by uneven angles. Experimental results show that the proposed method has an average relative error of 0.38%, which is approximately 0.26 lower than existing advanced methods.

7 December 2024 in Sensors
Verified
Journal Article
Open access
PDF file
FCFeng Chen
Feng Chen
BSBaishun Su
Baishun Su
ZJZongpu Jia
Zongpu Jia
Over the last few years, neural architecture search (NAS) technology has achieved good results in hyperspectral image classification. Nevertheless, existing NAS-based classification methods have not specifically focused on the complex connection between spectral and spatial data. Strengthening the integration of spatial and spectral features is crucial to boosting the overall classification efficacy of hyperspectral images. In this paper, a triple-unit hyperspectral NAS network (TUH-NAS) aimed at hyperspectral image classification is introduced, where the fusion unit emphasizes the enhancement of the intrinsic relationship between spatial and spectral information. We designed a new hyperspectral image attention mechanism module to increase the focus on critical regions and enhance sensitivity to priority areas. We also adopted a composite loss function to enhance the model’s focus on hard-to-classify samples. Experimental evaluations on three publicly accessible hyperspectral datasets demonstrated that, despite utilizing a limited number of samples, TUH-NAS outperforms existing NAS classification methods in recognizing object boundaries.

26 December 2025 in Discover Artificial Intelligence
Verified
Journal Article
Open access
PDF file
HLHui Li
Hui Li
XQXianglei Qiu
Xianglei Qiu
JZJunxia Zhang
Junxia Zhang
BZBin Zhang
Bin Zhang
BLBibo Lu
Bibo Lu
BSBaishun Su
Baishun Su
With the rapid advancement of informatization, tables have become a fundamental format for representing structured data in documents and are widely utilized across various domains, including healthcare, education, and finance. Table recognition aims to automatically convert table content from images into editable and machine-readable formats, significantly enhancing processing efficiency and accuracy. Despite significant advancements made by convolutional neural network (CNN)-based and Transformer-based approaches, existing methods still face challenges in handling complex table layouts, diverse formats, and high computational costs. To address these challenges, we propose the Spatial Pyramid Pooling?enhanced Multi-modal Linear Transformer (SPPLTR), a novel table-recognition model that integrates spatial pyramid pooling with a causal linear-Transformer framework. SPPLTR effectively captures both structural and textual information in table images through a spatial pyramid context fusion module (SPCF) and a causal linear Transformer, thereby improving the understanding of table structures. Additionally, the model incorporates a multi-head causal linear attention (MHCLA) mechanism with automatic mixed precision (AMP) support, enabling more efficient and stable inference. Extensive experiments conducted on the PubTabNet dataset demonstrate that SPPLTR achieves state-of-the-art (SOTA) performance, surpassing most existing models across various evaluation metrics. The proposed model offers a promising solution for table recognition tasks and lays the groundwork for further advancements in structured document understanding. The code for SPPLTR will be made publicly available at https://github.com/Zhanxin-Xin/SPPLTR.

1 January 2026 in IEEE Sensors Journal
Verified
Journal Article
Closed access
BSBaishun Su
Baishun Su
HLHao Li
Hao Li
MWMeng Wang
Meng Wang

13 June 2025 in Information
Verified
Journal Article
Open access
PDF file
FCFeng Chen
Feng Chen
BSBaishun Su
Baishun Su
ZJZongpu Jia
Zongpu Jia
The accurate identification of water bodies in hyperspectral images (HSIs) remains challenging due to hierarchical representation imbalances in deep learning models, where shallow layers overly focus on spectral features, boundary ambiguities caused by the relatively low spatial resolution of satellite imagery, and limited detection capability for small-scale aquatic features such as narrow rivers. To address these challenges, this study proposes Heuristic–Adaptive Spectral–Spatial Neural Architecture Search with Dynamic Cell Evaluation (HASSDE-NAS). The architecture integrates three specialized units; a spectral-aware dynamic band selection cell suppresses redundant spectral bands, while a geometry-enhanced edge attention cell refines fragmented spatial boundaries. Additionally, a bidirectional fusion alignment cell jointly optimizes spectral and spatial dependencies. A heuristic cell search algorithm optimizes the network architecture through architecture stability, feature diversity, and gradient sensitivity analysis, which improves search efficiency and model robustness. Evaluated on the Gaofen-5 datasets from the Guangdong and Henan regions, HASSDE-NAS achieves overall accuracies of 92.61% and 96%, respectively. This approach outperforms existing methods in delineating narrow river systems and resolving water bodies with weak spectral contrast under complex backgrounds, such as vegetation or cloud shadows. By adaptively prioritizing task-relevant features, the framework provides an interpretable solution for hydrological monitoring and advances neural architecture search in intelligent remote sensing.