Article
Environmental Sciences
Danpei Zhao, Chunbo Zhu, Jing Qi, Xinhu Qi, Zhenhua Su, Zhenwei Shi
Summary: The paper proposes a SAR ship instance segmentation method based on the synergistic attention mechanism, improving ship detection performance and providing pixel-level contours for subsequent applications. The method introduces a synergistic attention strategy at the image, semantic, and target level in the instance segmentation framework.
Article
Environmental Sciences
Jinsheng Xiao, Haowen Guo, Yuntao Yao, Shuhao Zhang, Jian Zhou, Zhijun Jiang
Summary: In this paper, a new method for image object detection is proposed, which improves the detection of small objects in complex backgrounds through multi-feature selection and fusion. The use of an attention mechanism-based multi-feature selection module reduces the interference of useless information and improves detection accuracy.
Article
Environmental Sciences
Xiaozhen Ren, Yanwen Bai, Gang Liu, Ping Zhang
Summary: This paper proposes an efficient lightweight network YOLO-Lite for SAR ship detection. It reduces calculation through a lightweight feature enhancement backbone, accurately locates target location by a channel and position enhancement attention module, enhances the expression ability of features with an enhanced spatial pyramid pooling module, constructs a multi-scale feature fusion network for more position and semantic information, and improves detection accuracy with a novel confidence loss function.
Article
Engineering, Electrical & Electronic
Shuai Liu, Pengfei Chen, Yudong Zhang
Summary: Synthetic aperture radar (SAR) ship detection is widely used in various applications, but existing lightweight algorithms have problems such as misjudgment of targets mixed with the background and poor detection performance for targets with few samples. To address these issues, this paper proposes a detection network based on a multiscale feature pyramid network (FPN) that enhances target features, suppresses interference from the background, and reduces misjudgment. Experimental results show that the proposed network outperforms existing algorithms on public datasets.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Computer Science, Hardware & Architecture
Weina Zhou, Yujie Peng
Summary: This paper proposes a real-time ship detection method based on YOLOv5, which achieves high detection accuracy with few parameters and little memory and computation cost. The method incorporates Collaborative Attention mechanism, Spatial Pyramid Pooling module, Bidirectional Feature Pyramid Network, and Transformer encoder to improve performance. Experimental results show that the proposed method outperforms existing technology in ship detection.
Article
Geochemistry & Geophysics
Mingfeng Zh, Wenbin Qian, Wenji Yang, Yilu Xu
Summary: This letter proposes a novel ship detection model based on multifeature transformation and fusion (MFTF-Net) to address the issues of high false alarm detection rate and prone missed detection in SAR image ship detection. The proposed model utilizes anchor frame clustering, local enhancement network, improved transformer structure, and four-scale residual feature fusion network to achieve better performance compared to 13 baseline models.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Environmental Sciences
Boli Xiong, Zhongzhen Sun, Jin Wang, Xiangguang Leng, Kefeng Ji
Summary: Detection and recognition of SAR ships play a vital role in SAR data interpretation. Most existing research focuses on target detection with limited studies on integrated detection and recognition in complex SAR images. In this paper, a lightweight model is proposed, which improves detection performance and deployment capability by optimizing feature extraction and prediction layer structure. Extensive experiments show that the proposed model achieves excellent performance in ship detection and recognition in complex SAR images.
Article
Environmental Sciences
Peng Chen, Hui Zhou, Ying Li, Peng Liu, Bingxin Liu
Summary: In this paper, we propose a novel deep learning network with deformable convolution and attention mechanisms to improve the Feature Pyramid Network (FPN) model for nearshore ship target detection in SAR images with complex backgrounds. The experimental results show that our model achieves a detection accuracy of 87.9% for complex scenes and 95.1% for small-scale ship targets.
Article
Computer Science, Artificial Intelligence
Kai Zhao, Ruitao Lu, Siyu Wang, Xiaogang Yang, Qingge Li, Jiwei Fan
Summary: This paper proposes a novel SAR ship detection model called ST-YOLOA, which incorporates the Swin Transformer network architecture and coordinate attention (CA) model in the STCNet backbone network to enhance feature extraction and global information capture. The PANet path aggregation network and a novel up/down-sampling method are used to address feature pyramid construction and local interference and semantic information loss. The decoupled detection head improves convergence speed and detection accuracy. The experimental results demonstrate the superior performance of ST-YOLOA compared to other state-of-the-art methods.
FRONTIERS IN NEUROROBOTICS
(2023)
Article
Engineering, Mechanical
Zhenyu Han, Yue Zhuo, Yizhao Yan, Hongyu Jin, Hongya Fu
Summary: In this paper, a chatter detection method based on deep learning is proposed for the milling of thin-walled parts. By employing multi-channel signal features and attention mechanisms, the proposed method can accurately detect chatter during the milling process, which has been validated through experiments.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2022)
Article
Engineering, Mechanical
Runqiong Wang, Qinghua Song, Yezhen Peng, Jing Qin, Zhanqiang Liu, Zhaojun Liu
Summary: Cutting tool condition monitoring (TCM) techniques are important for optimizing production cost and machining quality in smart manufacturing. This paper proposes an autonomous TCM model that selects and fuses local-temporal features automatically, without relying on manual operation and domain knowledge. The model improves feature fusion performance by 9.48% on average compared to conventional methods and can be adapted to different cutting conditions with minimal computational resources.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2023)
Article
Environmental Sciences
Libo Xu, Chaoyi Pang, Yan Guo, Zhenyu Shu
Summary: The research introduced a novel network structure CF-SSD based on the original SSD framework to improve the detection accuracy of SAR ship images significantly while maintaining a fast inference speed. By incorporating three blocks including CF, GAM, and mixed loss function, CF-SSD effectively extracted and fused features, avoided loss of small or blurred object information, and precisely located object positions from SAR images. The experiments conducted on the SAR ship dataset SSDD demonstrated a 90.3% mAP and a fast inference speed close to the original SSD, with excellent detection performance and high efficiency on remote sensing datasets NWPU VHR-10 and VOC2007.
Article
Engineering, Electrical & Electronic
Weina Zhou, Lu Liu
Summary: This paper proposes an efficient multilayer attention receptive fusion network (MARN) based on YOLOv4 to solve the problem of multiscale ship detection in complex marine environments. MARN extracts semantic information from feature maps at different scales to highlight salient features of ships, thereby improving detection performance.
JOURNAL OF ELECTRONIC IMAGING
(2022)
Article
Agronomy
Liyang Su, Haixia Sun, Shujuan Zhang, Xinyuan Lu, Runrun Wang, Linjie Wang, Ning Wang
Summary: This study proposes a lightweight YOLOv5s-Super model based on the YOLOv5s model, which captures cucumber shoulder features and dynamically fuses multi-scale features in near-color backgrounds. The Ghost module is added to improve the inference time and floating-point computation speed of the model. Experimental results show that the YOLOv5s-Super model achieves an mAP of 87.5%, which is 4.2% higher than YOLOv7-tiny and 1.9% higher than YOLOv8s. The improved model accurately and robustly detects multi-scale features in complex backgrounds, meeting the requirement of being lightweight and providing technical support for intelligent cucumber picking.
Article
Geochemistry & Geophysics
Yaqi Han, Xinyi Yang, Tian Pu, Zhenming Peng
Summary: Remote sensing ship recognition has wide applications in civil and military fields. However, the lack of public datasets for fine-grained ship recognition has restricted the progress in this area. In this paper, we propose an efficient information reuse network (EIRNet) and establish a public dataset for oriented ship recognition (DOSR). Our approach addresses the challenges of complex scenes and ship characteristics by using a dense feature fusion network and a dual-mask attention module. Experimental results show that our method achieves state-of-the-art performance on DOSR and another popular dataset (HRSC2016).
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Geosciences, Multidisciplinary
Nan Xu, Yue Ma, Jian Yang, Xiao Hua Wang, Yongjun Wang, Rui Xu
Summary: This study presents a fusion method of Sentinel-2 and ICESat-2 datasets for deriving tidal flat topography, which was validated with high accuracy and without the need for in-situ data.
GEOPHYSICAL RESEARCH LETTERS
(2022)
Article
Geochemistry & Geophysics
Nan Xu, Yue Ma, Hui Zhou, Wenhao Zhang, Zhiyu Zhang, Xiao Hua Wang
Summary: This study proposes a method to derive high-resolution bathymetry for dynamic areas using satellite remotely sensed data sets. The method combines ICESat-2 lidar data and GSWD data and has been validated in Lake Mead, USA. It shows promise for obtaining global bathymetry in inland water bodies and coastal areas with strong water level fluctuations and sufficient water clarity.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Computer Science, Artificial Intelligence
Zhen Wang, Jianxin Guo, Shanwen Zhang
Summary: This paper proposes a deep neural network structure, MPF-Net, for weed species identification. By processing and extracting features of images at different scales and using a lightweight model and a scale reduction module, the proposed method achieves high identification accuracy.
INTERNATIONAL JOURNAL OF PATTERN RECOGNITION AND ARTIFICIAL INTELLIGENCE
(2022)
Article
Computer Science, Information Systems
Jiwei Tian, Buhong Wang, Rongxiao Guo, Zhen Wang, Kunrui Cao, Xiaodong Wang
Summary: This article explores the vulnerabilities of DL-based UAVs and proposes two adversarial attack methods against regression models in UAVs. The experiments demonstrate that these attack methods pose a considerable threat to the navigation and control of UAVs. To address this problem, adversarial training and defensive distillation methods are further investigated and evaluated, increasing the robustness of DL models in UAVs.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Article
Automation & Control Systems
Jiwei Tian, Buhong Wang, Zhen Wang, Kunrui Cao, Jing Li, Mete Ozay
Summary: The article introduces the joint adversarial example and FDIAs method for attacks on state estimation in power systems, ensuring stealthiness and adversarial robustness. Theoretical and experimental results show that the new method is more efficient and effective in terms of performance.
IEEE TRANSACTIONS ON CYBERNETICS
(2022)
Article
Environmental Studies
Li Li, Lianqi Zhu, Nan Xu, Ying Liang, Zhengyu Zhang, Junjie Liu, Xin Li
Summary: This study examined the response of different vegetation types in China's north-south transition zone to diurnal temperature changes. The results showed that daytime temperature increased at a rate 1.2 times that of nighttime temperature, and the effects of temperature changes varied among different vegetation types. Daytime warming promoted the growth of grasses, shrubs, deciduous broad-leaved forests, crops, and conifers, while nighttime warming had a positive effect only on the growth of evergreen broad-leaved forests.
Article
Environmental Sciences
Yue Ma, Lin Wang, Nan Xu, Shiyi Zhang, Xiao Hua Wang, Song Li
Summary: Coastal slope is a crucial factor in understanding hydrodynamic and morphological processes, estimating extreme water levels, coastline erosion, and coastal vulnerability. However, accurate estimation of coastal slopes is limited, especially in sparsely populated and remote areas. This study demonstrates the potential of ICESat-2 measurements to estimate coastal slope, with promising results obtained in Texas, USA. The improved accuracy of ICESat-2 derived coastal slopes compared to current large-scale slopes derived from other sources suggests the method's applicability at a global scale.
ENVIRONMENTAL RESEARCH LETTERS
(2023)
Article
Engineering, Electrical & Electronic
Jinwen Tian, Guozhi Zhang, Chuan Ming, Luyao He, Yang Liu, Jianben Liu, Xiaoxing Zhang
Summary: This letter presents the design of a new flexible Hilbert antenna for built-in detection of partial discharges (PD) in gas-insulated switchgear (GIS). The fractal antenna is designed on flexible polydimethylsiloxane (PDMS) material and demonstrates high PD detection sensitivity. It has a broad bandwidth and little effect on antenna bandwidth even with bending deformation.
IEEE ANTENNAS AND WIRELESS PROPAGATION LETTERS
(2023)
Article
Computer Science, Artificial Intelligence
Zheng-Yang Zhao, Jie Lin, Zhen Wang, Jian-Xin Guo, Xin-Ke Zhan, Yu-An Huang, Chuan Shi, Wen-Zhun Huang
Summary: This paper proposed a semantic embedded bipartite graph network (SEBGLMA) for predicting the association between long noncoding RNA (lncRNA) and micro-RNA (miRNA), which integrated K-mer segmentation, word2vec, Gaussian interaction profile (GIP), and graph convolution network (GCN) for feature extraction. The results showed that the model achieved high accuracy, precision, sensitivity, specificity, Matthews correlation coefficient, and F1-Score on the benchmark dataset.
INTERNATIONAL JOURNAL OF INTELLIGENT SYSTEMS
(2023)
Article
Environmental Sciences
Zhen Wang, Buhong Wang, Chuanlei Zhang, Yaohui Liu
Summary: This article systematically evaluates the threat of adversarial patches on the aerial image semantic segmentation task for the first time. To defend against adversarial patch attacks and obtain accurate semantic segmentation results, a novel robust feature extraction network (RFENet) is constructed. RFENet designs limited receptive field mechanism (LRFM), spatial semantic enhancement module (SSEM), boundary feature perception module (BFPM), and global correlation encoder module (GCEM) to solve adversarial patch attacks from the DL model architecture design level.
Article
Environmental Sciences
Zhen Wang, Buhong Wang, Yaohui Liu, Jianxin Guo
Summary: This article assesses and verifies the influence of adversarial attacks on aerial image semantic segmentation and proposes a novel global feature attention network (GFANet) to solve the threat of adversarial attacks. GFANet uses the global context encoder (GCE) to obtain the context dependencies of global features, introduces the global coordinate attention mechanism (GCAM) to enhance the global feature representation to suppress adversarial noise, and the feature consistency alignment (FCA) is used for feature calibration. In addition, a universal adversarial training strategy is constructed to improve the robustness of the semantic segmentation model against adversarial example attacks. Extensive experiments on three aerial image datasets demonstrate that GFANet is more robust against adversarial attacks than existing state-of-the-art semantic segmentation models.
Article
Geochemistry & Geophysics
Zhen Wang, Shanwen Zhang, Chuanlei Zhang, Buhong Wang
Summary: This article proposes a novel hidden feature-guided semantic segmentation network (HFGNet) for accurate semantic segmentation of remote sensing images. It introduces a hidden feature extraction module (HFE-M) to mine more valuable hidden features and a multifeature interactive fusion module (MIF-M) to establish correlation between different features. A multiscale feature calibration module (MSFC) and a local-channel attention mechanism (LCA-M) are also designed to enhance the diversity and refinement representation of hierarchical fusion features. Experimental results show that the proposed HFGNet outperforms several state-of-the-art methods.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Environmental Sciences
Dongzhen Jia, Yu Li, Xiufeng He, Zhixiang Yang, Yihao Wu, Taixia Wu, Nan Xu
Summary: This study proposed a method for selecting and correcting optical deep-water areas to improve the accuracy of satellite-derived bathymetry (SDB). The results showed that applying sun glint correction significantly improved the quality of SDB, and the computed results from different deep-water areas were more consistent.
Article
Remote Sensing
Conghong Huang, Nan Xu
Summary: This study maps the spatial patterns of urban green space coverage (UGSC) in all urban areas of the contiguous US and analyzes the main influencing factors. The results show that climatic factors play a major role in shaping the UGSC pattern, while city size and terrain factors have a relatively smaller impact.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2022)
Article
Geochemistry & Geophysics
Zhen Wang, Nan Xu, Jianxin Guo, Chuanlei Zhang, Buhong Wang
Summary: In this article, we propose a semantic condition constraint guided feature aware network (SCFNet) for detecting different aircraft categories in SAR images. The SCFNet effectively extracts fine-grained features and suppresses interference through the design of feature aware modules, feature fusion pyramid, and global coordinate attention mechanism. The experimental results demonstrate that SCFNet achieves state-of-the-art performance in aircraft detection.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
