Sun Long

Mr. Sun Long


Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

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

Representative Publications

View All Publications
13 December 2021 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Verified
Journal Article
Open access
HSHui Shao
Hui Shao
ZCZheng Cao
Zheng Cao
WLWei Li
Wei Li
Yuwei ChenYuwei Chen
Yuwei Chen
Changhui JiangChanghui Jiang
Changhui Jiang
SLSun Long
Sun Long
Wood–leaf separation aiming at classifying LiDAR points into wood and leaf components is one of the most important genres for improving leaf area index estimation and forestry survey accuracy. The wood return signals could artificially increase the apparent foliage content, which needs to be screened out for deriving vital tree attributes accurately. Previous research works tended to use intensity, waveform, and geometric information extracted from a single wavelength LiDAR for wood–leaf separation. This article employs a revised hyperspectral LiDAR (HSL) to obtain spatial and ultrawide spectral data simultaneously. We also propose a simple three steps method to separate wood and leaf components based on HSL spatial and spectral measurements under the laboratory circumstances. First, the preprocessing is conducted to acquire 3-D spatial information and the multiband laser pulse reflectance for further separation. Second, preliminary separation (band division, key feature parameter calculation, and judgment) is implemented based on reflectance. Third, we employ K-Nearest Neighbor (KNN) method to enhance separation results based on preliminary separation results and spatial features and then update the results by recorrection. Then, 3-D reconstruction is accomplished by fusing wood–leaf separation results. The experimental results demonstrate that the proposed method can separate wood and leaf components with high accuracy and indicate tree attributes straightforwardly.

26 August 2024 in IEEE Sensors Journal
Verified
Journal Article
Closed access
HXHeng Xu
Heng Xu
FMFanyu Meng
Fanyu Meng
HLHu Liu
Hu Liu
HSHui Shao
Hui Shao
SLSun Long
Sun Long
With the widespread adoption of smart mobile devices, smartphone-based indoor positioning technology has become a significant area of research. This study proposes a collaborative adaptive multi-source data fusion method for indoor positioning, utilizing the inherent multi-sensor capabilities of smartphones. In the offline phase, a Multi-Layer Perceptron (MLP) is trained using collected Wi-Fi fingerprint data. Specifically, we recorded 114 fingerprint locations in Scenario 1 and 99 in Scenario 2, each containing 20,000 fingerprint data sets. During the online phase, the system collects real-time Wi-Fi fingerprint data and matches it against the trained dataset. Subsequently, the user’s step length and direction data are integrated into a Pedestrian Dead Reckoning (PDR) algorithm to estimate positions within complex indoor environments. Finally, we apply an Adaptive Extended Kalman Filter (AEKF), which adjusts weights to further improve positioning accuracy. Experimental results demonstrate that the AEKF method reduces root-mean-square errors (RMSE) to 0.763m and 0.884m in scenarios with different complexities, enhancing accuracy by 10.13% and 30.17% compared to traditional EKF methods. Comparative analysis with mainstream algorithms further highlights the advantages of AEKF, achieving maximum accuracy improvements of up to 36.54%. These findings underscore the effectiveness of our approach in advancing indoor positioning technology for mobile applications, offering practicality and cost-effectiveness without requiring additional hardware.

20 March 2023 in Sensors
Verified
Journal Article
Open access
PDF file
HSHui Shao
Hui Shao
FWfuyu wang
fuyu wang
WLWei Li
Wei Li
PHPeilun Hu
Peilun Hu
SLSun Long
Sun Long
Yuwei ChenYuwei Chen
Yuwei Chen
Intelligent management of trees is essential for precise production management in orchards. Extracting components’ information from individual fruit trees is critical for analyzing and understanding their general growth. This study proposes a method to classify persimmon tree components based on hyperspectral LiDAR data. We extracted nine spectral feature parameters from the colorful point cloud data and performed preliminary classification using random forest, support vector machine, and backpropagation neural network methods. However, the misclassification of edge points with spectral information reduced the accuracy of the classification. To address this, we introduced a reprogramming strategy by fusing spatial constraints with spectral information, which increased the overall classification accuracy by 6.55%. We completed a 3D reconstruction of classification results in spatial coordinates. The proposed method is sensitive to edge points and shows excellent performance for classifying persimmon tree components.

1 January 2026 in Engineering Research Express
Verified
Journal Article
Closed access
HXHeng Xu
Heng Xu
DHDie Hu
Die Hu
HSHui Shao
Hui Shao
SLSun Long
Sun Long
JCJuan Chen
Juan Chen
Non-contact vital signs detection demonstrates significant potential in fields like fire rescue, intelligent healthcare, and smart homes. However, existing methods encounter challenges in extracting accurate signals in noisy and low signal-to-noise ratio (SNR) indoor scenarios and exhibit heavy reliance on empirical parameter tuning. This study proposes a multi-stage optimized method for extracting heart and respiratory signals from Ultra-wideband (UWB) radar by integrating the autocorrelation function (AF), whale optimization algorithm (WOA), and variational mode decomposition (VMD). First, direct current (DC) removal, singular value decomposition (SVD), and Butterworth filtering are applied to suppress noise and eliminate baseline drift. Then, autocorrelation analysis of the slow-time radar signal is used to leverage the periodicity of vital signs for accurate target distance detection, improving signal extraction accuracy. Finally, WOA dynamically optimizes VMD parameters to adapt to time-varying signals, enhance mode separation and noise resistance, and screen effective components based on modal energy contribution. Vital signs frequencies are subsequently extracted using Fast Fourier transform (FFT). Experiments using 180 datasets from three volunteers at four distances (0.8 m–2.4 m) show that the proposed method achieves average errors of 4.84% in distance, 2.52% in respiratory rate (RR) and 3.54% in heart rate (HR) under static indoor line-of-sight conditions. These results indicate that the method proposed in this paper can provide accurate and relatively stable non-contact vital signs estimation within the test distance range.

4 December 2024 in Research Square
Verified
Preprint
Open access
PDF file
SLSun Long
Sun Long
TLTantan Li
Tantan Li
HSHui Shao
Hui Shao
FMFanyu Meng
Fanyu Meng
With the rapid development of intelligent communication technology, modulation recognition of signals has become a crucial research topic in multiple domains such as radar, communications, and electronic warfare. Currently, deep learning-based modulation recognition methods leverage their powerful feature learning capabilities to effectively improve recognition accuracy in low signal-to-noise ratio environments. However, these methods often suffer from complex network structures, high computational requirements, and demanding hardware platforms. Addressing these issues, this paper proposes an automatic modulation recognition method based on lightweight networks applicable under edge conditions. Firstly, the original I/Q data undergo wavelet threshold denoising to mitigate the impact of noise on signal modulation recognition. Subsequently, by incorporating a phase estimator and an enhanced channel attention mechanism into the model, this approach accurately captures signal features affected by noise and interference, while endowing the network with the ability to learn and focus on key information. Importantly, it replaces complex and redundant network structures by utilizing only three one-dimensional convolutional layers for feature extraction, thereby achieving simplification and optimization of the network architecture.After validation on the public dataset RadioML 2016.10a, the number of parameters of this model is only 0.06M, but the average recognition rate reaches 62.32%.