Haisong Chen

Dr. Haisong Chen


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
27 June 2025 in Electronics
Verified
Journal Article
Open access
PDF file
aili wangaili wang
aili wang
FZFu Zhijia
Fu Zhijia
YZYanran Zhao
Yanran Zhao
HCHaisong Chen
Haisong Chen
Due to the challenges posed by high resolution, substantial background noise, significant object scale variation, and long-tailed data distribution in remote sensing images, traditional techniques often struggle to maintain both high accuracy and low latency. This paper proposes YOLO11-FSDAT, an advanced object detection framework tailored for remote sensing imagery, which integrates not only modular enhancements but also theoretical and architectural innovations to address these limitations. First, we propose the frequency–spatial feature extraction fusion module (Freq-SpaFEFM), which breaks the conventional paradigm of spatial-domain-dominated feature learning by introducing a multi-branch architecture that fuses frequency- and spatial-domain features in parallel. This design provides a new processing paradigm for multi-scale object detection, particularly enhancing the model’s capability in handling dense and small-object scenarios with complex backgrounds. Second, we introduce the deformable attention-based global–local fusion module (DAGLF), which combines fine-grained local features with global context through deformable attention and residual connections. This enables the model to adaptively capture irregularly oriented objects (e.g., tilted aircraft) and effectively mitigates the issue of information dilution in deep networks. Third, we develop the adaptive threshold focal loss (ATFL), which is the first loss function to systematically address the long-tailed distribution in remote sensing datasets by dynamically adjusting focus based on sample difficulty. Unlike traditional focal loss with fixed hyperparameters, ATFL decouples hard and easy samples and automatically adapts to varying class distributions. Experimental results on the public DOTAv1, SIMD, and DIOR datasets demonstrated that YOLO11-FSDAT achieved 75.22%, 82.79%, and 88.01% mAP, respectively, outperforming baseline YOLOv11n by up to 4.11%. These results confirm the effectiveness, robustness, and broader theoretical value of the proposed framework in addressing key challenges in remote sensing object detection.

28 November 2025 in Sensors
Verified
Journal Article
Open access
PDF file
HCHaisong Chen
Haisong Chen
YPYuan Pengfei
Yuan Pengfei
WLWenbai Liu
Wenbai Liu
FLFuling and classifying activities in robot-assisted surgery with recurrent neural Li
Fuling and classifying activities in robot-assisted surgery with recurrent neural Li
aili wangaili wang
aili wang
To address the limitations of unimodal visual detection in complex scenarios involving low illumination, occlusion, and texture-sparse environments, this paper proposes an improved YOLOv11-based dual-branch RGB-D fusion framework. The symmetric architecture processes RGB images and depth maps in parallel, integrating a Dual-Encoder Cross-Attention (DECA) module for cross-modal feature weighting and a Dual-Encoder Feature Aggregation (DEPA) module for hierarchical fusion—where the RGB branch captures texture semantics while the depth branch extracts geometric priors. To comprehensively validate the effectiveness and generalization capability of the proposed framework, we designed a multi-stage evaluation strategy leveraging complementary benchmark datasets. On the M3FD dataset, the model was evaluated under both RGB-depth and RGB-infrared configurations to verify core fusion performance and extensibility to diverse modalities. Additionally, the VOC2007 dataset was augmented with pseudo-depth maps generated by Depth Anything, assessing adaptability under monocular input constraints. Experimental results demonstrate that our method achieves mAP50 scores of 82.59% on VOC2007 and 81.14% on M3FD in RGB-infrared mode, outperforming the baseline YOLOv11 by 5.06% and 9.15%, respectively. Notably, in the RGB-depth configuration on M3FD, the model attains a mAP50 of 77.37% with precision of 88.91%, highlighting its robustness in geometric-aware detection tasks. Ablation studies confirm the critical roles of the Dynamic Branch Enhancement (DBE) module in adaptive feature calibration and the Dual-Encoder Attention (DEA) mechanism in multi-scale fusion, significantly enhancing detection stability under challenging conditions. With only 2.47M parameters, the framework provides an efficient and scalable solution for high-precision spatial perception in autonomous driving and robotics applications.

11 December 2024 in Electronics
Verified
Journal Article
Open access
PDF file
aili wangaili wang
aili wang
FYFeng Yanxiang
Feng Yanxiang
MYMingji Yang
Mingji Yang
Haibin WuHaibin Wu
Haibin Wu
Yuji IwahoriYuji Iwahori
Yuji Iwahori
HCHaisong Chen
Haisong Chen
Cross-project software defect prediction (CPDP) refers to the construction of defect prediction models by collecting multi-source project data, but the heterogeneity of data among projects and the modern problem of “data islands” hinder its development. In response to these challenges, we propose a CPDP algorithm based on differential perception combined with inheritance federated learning (FedDPI). Firstly, we design an efficient data preprocessing scheme, which lays a reliable data foundation for federated learning by integrating oversampling and optimal feature selection methods. Secondly, a two-stage collaborative optimization mechanism is proposed in the federated learning framework: the inheritance private model (IPM) is introduced in the local training stage, and the differential perception algorithm is used in the global aggregation stage to dynamically allocate aggregation weights, forming positive feedback for training to overcome the negative impact of data heterogeneity. In addition, we utilize the Ranger optimization algorithm to improve the convergence speed and privacy security of the model through its irreversible mixed optimization operation. The experimental results show that FedDPI significantly improves predictive performance in various defect item data combination experiments. Compared with different deep learning and federated learning algorithms, the average improvement in AUC and G-mean indicators is 0.2783 and 0.2673, respectively, verifying the practicality and effectiveness of federated learning and two-stage collaborative optimization mechanisms in the field of CPDP.

24 July 2025 in Electronics
Verified
Journal Article
Open access
PDF file
aili wangaili wang
aili wang
KZKang Zhang
Kang Zhang
Haibin WuHaibin Wu
Haibin Wu
HCHaisong Chen
Haisong Chen
MWMinhui Wang
Minhui Wang
In order to address the limitations of the number of label samples in practical accurate classification scenarios and the problems of overfitting and an insufficient generalization ability caused by Few-Shot Learning (FSL) in hyperspectral image classification (HSIC), this paper designs and implements a neural architecture search (NAS) for a few-shot HSI classification method that combines meta learning. Firstly, a multi-source domain learning framework was constructed to integrate heterogeneous natural images and homogeneous remote sensing images to improve the information breadth of few-sample learning, enabling the final network to enhance its generalization ability under limited labeled samples by learning the similarity between different data sources. Secondly, by constructing precise and robust search spaces and deploying different units at different locations, the classification accuracy and model transfer robustness of the final network can be improved. This method fully utilizes spatial texture information and rich category information of multi-source data and transfers the learned meta knowledge to the optimal architecture for HSIC execution through precise and robust search space design, achieving HSIC tasks with limited samples. Experimental results have shown that our proposed method achieved an overall accuracy (OA) of 98.57%, 78.39%, and 98.74% for classification on the Pavia Center, Indian Pine, and WHU-Hi-LongKou datasets, respectively. It is fully demonstrated that utilizing spatial texture information and rich category information of multi-source data, and through precise and robust search space design, the learned meta knowledge is fully transmitted to the optimal architecture for HSIC, perfectly achieving classification tasks with few-shot samples.

15 May 2026 in Remote Sensing
Verified
Journal Article
Open access
PDF file
aili wangaili wang
aili wang
XLXinyu Liu
Xinyu Liu
HCHaisong Chen
Haisong Chen
Hyperspectral image classification (HSIC) is a crucial task for remote sensing applications, requiring accurate pixel-level labeling while effectively capturing both spectral and spatial information. Traditional convolutional neural network architectures often struggle to balance local texture detail and global contextual consistency, and existing neural architecture search (NAS) methods rarely incorporate attention mechanisms, limiting their performance. To address these challenges, this study proposes a multi-scale Transformer-based NAS framework (TR-NAS) for fine-grained hyperspectral image classification. The framework combines local cube sampling, shallow and deep multi-scale convolutions, and a searchable Transformer module that adaptively selects global, local window, and multi-scale attention operators. Lightweight enhanced convolution operators, including dual-gated (DG-Conv) and mixed depthwise (MixConv) convolutions, are incorporated to improve spectral discrimination and scale robustness. Extensive experiments on the PU and Hanchuan datasets demonstrate that TR-NAS achieves superior classification accuracy, stability, and boundary consistency compared to traditional methods and existing NAS architectures, showing improved robustness to spectral similarity and spatial heterogeneity in complex remote sensing scenes.

20 June 2026 in Sensors
Verified
Journal Article
Open access
PDF file
YZYuan Zheng
Yuan Zheng
HCHaisong Chen
Haisong Chen
HWHuan Wan
Huan Wan
YWYongxue Wang
Yongxue Wang
This paper investigates backscatter-aided relaying for interactive dual-HAP wireless-powered sensor networks (WPSNs), in which two cooperative sensor groups transmit sensed data to opposite hybrid access points (HAPs) using harvested radio-frequency energy. Each group consists of multiple source sensor nodes (SNs) and one relay SN selected according to its proximity to the target HAP. To reduce local cooperation overhead, source SNs reuse the wireless power transfer (WPT) signal as a controllable carrier and convey their information to the relay SN through passive backscatter communication. The collected information is then delivered to the target HAPs through direct source transmission and relay forwarding. A source common-throughput maximization problem is formulated by jointly optimizing time allocation, transmit energy allocation, and dual-HAP energy beamforming, subject to energy-causality and relay minimum-rate constraints. To address the resulting non-convexity, an alternating optimization algorithm is developed, where the time-and-energy allocation subproblem is transformed into a convex form and the energy beamforming matrices are updated through energy-feasibility margin maximization. Numerical results show that the proposed scheme outperforms active cooperation without backscatter and direct transmission, demonstrating the effectiveness of integrating passive local information collection, relay-assisted uplink transmission, and optimized dual-HAP WPT.