Xinwei Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Xinwei Zhang has a total of 5 co-authors. Here are the top co-authors by number of shared publications:
- Bojian Yu
School of Marine Engineering Equipment, Zhejiang Ocean University, Zhoushan 316022, China
2 shared publications - Hanxiang Zhou
Jurong Petroleum and Natural Gas University Enterprise Joint Laboratory, Zhejiang Ocean University, Zhoushan 316022, China
1 shared publication - Hongwei Zhao
China University of Petroleum, Qingdao, China
1 shared publication - Jian-Xiong Zhang
Jurong Petroleum and Natural Gas University Enterprise Joint Laboratory, Zhejiang Ocean University, Zhoushan 316022, China
1 shared publication - Rongsheng Lin
Xinjiang Jurong Energy (Group) Co., Ltd., Urumqi 841603, China
1 shared publication
Latest Publications
- Open Access
Physics-Informed Side-Scan Sonar Perception: Tackling Weak Targets and Sparse Debris via Geometric and Frequency Decoupling
Side-scan sonar (SSS) serves as the primary perceptual instrument for Autonomous Underwater Vehicles (AUVs) in large-scale marine search and rescue (SAR) operations. However, the detection of critical targets is frequently hindered by severe hydro-acoustic noise, the spatial discontinuity of wreckage, and the weak visual signatures of small targets. To surmount these challenges, this paper presents WPG-DetNet. First, we introduce a Wavelet-Embedded Residual Backbone (WERB) to reconstruct the conventional downsampling paradigm. By substituting standard pooling with the Discrete Wavelet Transform (DWT), this architecture explicitly disentangles high-frequency noise from structural information in the frequency domain, thereby achieving the adaptive preservation of edge fidelity for large human-made targets while filtering out speckle interference. Then, addressing the distinct challenge of discontinuous aircraft wreckage, the framework further incorporates a Debris Graph Reasoning Module (D-GRM). This module models scattered fragments as nodes in a topological graph to capture long-range semantic dependencies, transforming isolated instance recognition into context-aware scene understanding. Finally, to bridge the gap between AI and underwater physics, we design a Shadow-Aided Decoupling Head (SADH) equipped with a physics-informed geometric loss. By enforcing mathematical consistency between target height and acoustic shadow length, this mechanism establishes a rigorous discriminative criterion capable of distinguishing weak-echo human bodies from seabed rocks based on shadow geometry. Experiments on the SCTD dataset demonstrate that WPG-DetNet achieves a mean Average Precision () of 97.5% and a Recall of 96.9%. Quantitative analysis reveals that our framework outperforms the classic Faster R-CNN by a margin of 12.8% in and surpasses the Transformer-based RT-DETR-R18 by 5.6% in high-precision localization metrics (). Simultaneously, WPG-DetNet maintains superior efficiency with an inference speed of 62.5 FPS and a lightweight parameter count of 16.8 M, striking an optimal balance between robust perception and the real-time constraints of AUV operations.Open AccessPhysics-Informed Side-Scan Sonar Perception: Tackling Weak Targets and Sparse Debris via Geometric and Frequency Decoupling
Side-scan sonar (SSS) serves as the primary perceptual instrument for Autonomous Underwater Vehicles (AUVs) in large-scale marine search and rescue (SAR) operations. However, the detection of critical targets is frequently hindered by severe hydro-acoustic noise, the spatial discontinuity of wreckage, and the weak visual signatures of small targets. To surmount these challenges, this paper presents WPG-DetNet. First, we introduce a Wavelet-Embedded Residual Backbone (WERB) to reconstruct the conventional downsampling paradigm. By substituting standard pooling with the Discrete Wavelet Transform (DWT), this architecture explicitly disentangles high-frequency noise from structural information in the frequency domain, thereby achieving the adaptive preservation of edge fidelity for large human-made targets while filtering out speckle interference. Then, addressing the distinct challenge of discontinuous aircraft wreckage, the framework further incorporates a Debris Graph Reasoning Module (D-GRM). This module models scattered fragments as nodes in a topological graph to capture long-range semantic dependencies, transforming isolated instance recognition into context-aware scene understanding. Finally, to bridge the gap between AI and underwater physics, we design a Shadow-Aided Decoupling Head (SADH) equipped with a physics-informed geometric loss. By enforcing mathematical consistency between target height and acoustic shadow length, this mechanism establishes a rigorous discriminative criterion capable of distinguishing weak-echo human bodies from seabed rocks based on shadow geometry. Experiments on the SCTD dataset demonstrate that WPG-DetNet achieves a mean Average Precision () of 97.5% and a Recall of 96.9%. Quantitative analysis reveals that our framework outperforms the classic Faster R-CNN by a margin of 12.8% in and surpasses the Transformer-based RT-DETR-R18 by 5.6% in high-precision localization metrics (). Simultaneously, WPG-DetNet maintains superior efficiency with an inference speed of 62.5 FPS and a lightweight parameter count of 16.8 M, striking an optimal balance between robust perception and the real-time constraints of AUV operations. - Open Access
BlueberryNet: A Lightweight CNN for Real-Time Ripeness Detection in Automated Blueberry Processing Systems
Blueberries are valued for their flavor and health benefits, but inconsistent ripeness at harvest complicates post-harvest food processing such as sorting and quality control. To address this, we propose a lightweight convolutional neural network (CNN) to detect blueberry ripeness in complex field environments, supporting efficient and automated food processing workflows. To meet the low-power and low-resource demands of embedded systems used in smart processing lines, we introduce a Grouped Large Kernel Reparameterization (GLKRep) module. This design reduces computational cost while enhancing the model’s ability to recognize ripe blueberries under complex lighting and background conditions. We also propose a Unified Adaptive Multi-Scale Fusion (UMSF) detection head that adaptively integrates multi-scale features using a dynamic receptive field. This enables the model to detect blueberries of various sizes accurately, a common challenge in real-world harvests. During training, a Semantics-Aware IoU (SAIoU) loss function is used to improve the alignment between predicted and ground truth regions by emphasizing semantic consistency. The model achieves 98.1% accuracy with only 2.6M parameters, outperforming existing methods. Its high accuracy, compact size, and low computational load make it suitable for real-time deployment in embedded sorting and grading systems, bridging field detection and downstream food-processing tasks.
Open AccessBlueberryNet: A Lightweight CNN for Real-Time Ripeness Detection in Automated Blueberry Processing Systems
Blueberries are valued for their flavor and health benefits, but inconsistent ripeness at harvest complicates post-harvest food processing such as sorting and quality control. To address this, we propose a lightweight convolutional neural network (CNN) to detect blueberry ripeness in complex field environments, supporting efficient and automated food processing workflows. To meet the low-power and low-resource demands of embedded systems used in smart processing lines, we introduce a Grouped Large Kernel Reparameterization (GLKRep) module. This design reduces computational cost while enhancing the model’s ability to recognize ripe blueberries under complex lighting and background conditions. We also propose a Unified Adaptive Multi-Scale Fusion (UMSF) detection head that adaptively integrates multi-scale features using a dynamic receptive field. This enables the model to detect blueberries of various sizes accurately, a common challenge in real-world harvests. During training, a Semantics-Aware IoU (SAIoU) loss function is used to improve the alignment between predicted and ground truth regions by emphasizing semantic consistency. The model achieves 98.1% accuracy with only 2.6M parameters, outperforming existing methods. Its high accuracy, compact size, and low computational load make it suitable for real-time deployment in embedded sorting and grading systems, bridging field detection and downstream food-processing tasks.