Article
Engineering, Electrical & Electronic
Yuzhen Niu, Zhihua Lin, Wenxi Liu, Wenzhong Guo
Summary: Taking photos of digital screens often results in color-distorting moire patterns. Existing methods neglect the complementarity among different resolutions. This paper proposes a novel moire removal model that exploits complementarity among resolutions and includes a residual moire removal block with local color correction structure. Experimental results show that the proposed method outperforms state-of-the-art methods with fewer parameters and computation overhead. The network framework also achieves state-of-the-art performance in rain removal task.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Computer Science, Artificial Intelligence
Lianping Yang, Hongliang Zhang, Panpan Wei, Yubo Sun, Xiangde Zhang
Summary: The paper introduces a network structure based on densely connected encoder-decoders to balance the number of parameters and localization results, improving the performance of depthwise convolution. By enhancing the loss function, better localization results were achieved in face alignment experiments, with a small model size and superior performance.
APPLIED INTELLIGENCE
(2021)
Article
Plant Sciences
Chang Song, Bo Peng, Huanyue Wang, Yuhong Zhou, Lei Sun, Xuesong Suo, Xiaofei Fan
Summary: This study proposed a seed-quality evaluation method using an improved Inception-ResNet network to rapidly and accurately identify corn seeds of different qualities based on appearance. The method involved image segmentation, replacement of standard convolution, application of attention mechanism, and feature fusion strategy. The results showed that the method had decent performance in detecting corn seed appearance quality, with high accuracy, precision, recall rate, F1 value, and reasonable detection time. It provided a theoretical basis and technical support for the construction of intelligent seed sorting equipment.
FRONTIERS IN PLANT SCIENCE
(2023)
Article
Engineering, Electrical & Electronic
Ming Gao, Pengjiang Qian
Summary: This study proposes a convolutional neural network algorithm based on a multilayer network model, which uses Exponential Linear Units-guided Depthwise Separable Convolution to extract spectral and spatial features from hyperspectral images. The feature maps are then fed into a cross-attention mechanism for weight allocation to optimize local feature information and computational resource allocation. Experimental results show that this network model achieves accurate hyperspectral image classification with improved computing efficiency.
Article
Environmental Sciences
Yu Sun, Liang Huang, Junsan Zhao, Xiaoxiang Li, Mulan Qiu
Summary: This study proposes a deep learning-based method for bridge detection, which utilizes a depth-wise separable multiscale feature fusion network. The experimental results demonstrate that the proposed method achieves high accuracy and speed, making it suitable for high-precision and fast bridge detection tasks.
GEOCARTO INTERNATIONAL
(2022)
Article
Computer Science, Information Systems
Zhongyong Wang, Dongzhe Sun, Kexian Gong, Wei Wang, Peng Sun
Summary: This paper proposes a lightweight convolutional neural network for automatic modulation classification task, with a focus on reducing model complexity by designing depthwise separable convolution residual architecture and using global depthwise convolution for feature reconstruction. Experimental results show significant savings in model parameters and inference time compared to recent works.
Article
Computer Science, Information Systems
Ying Jia, Hang Li, Jie Fang, Xin Chen, Liqi Ji, Nianyi Wang
Summary: Thangka murals in Tibet hold significant cultural value and their restoration is crucial in preserving Tibet's cultural heritage. The potential of Gated Convolution for Thangka mural restoration is significant, but existing approaches face challenges in terms of blurring at the pixel level and dependency on the perceptual field of the convolution. To address these challenges, we propose a novel Thangka mural inpainting method incorporating edge-assisted feature fusion and self attention based local refine network, which has been shown to effectively repair broken areas of Thangka murals.
Article
Computer Science, Information Systems
Pengcheng Li, Shan Gai
Summary: In this work, we proposed MSCIANet, an end-to-end multi-scale context information and attention network for deraining. The network utilizes multi-scale feature extraction and multi-receptive fields feature extraction to remove rain streaks from images, and achieves good results on synthetic and real-world datasets.
JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION
(2023)
Article
Mathematical & Computational Biology
Zhigao Zeng, Cheng Huang, Wenqiu Zhu, Zhiqiang Wen, Xinpan Yuan
Summary: In this paper, a new lightweight neural network model based on multi-scale feature fusion and attention mechanism is proposed to solve the problem of feature loss and storage space in deep learning-based flower image classification methods. The model, built upon the AlexNet framework, incorporates a multi-scale feature fusion module (MFFM) to replace single-scale convolution and enhances deep feature extraction with improved Inception and attention modules. The experimental results show that the lightweight model outperforms the baseline model in terms of parameters, storage space, and classification accuracy, making it suitable for accurate flower image recognition on mobile devices.
MATHEMATICAL BIOSCIENCES AND ENGINEERING
(2023)
Article
Computer Science, Information Systems
Sk Mahmudul Hassan, Arnab Kumar Maji
Summary: The paper proposes a novel deep learning model for plant disease identification, incorporating depthwise separable convolution to reduce training time and parameter count. The model was tested on multiple datasets and performed better than other deep learning models in terms of accuracy.
Article
Chemistry, Multidisciplinary
Qunyan Jiang, Juying Dai, Ting Rui, Faming Shao, Ruizhe Hu, Yinan Du, Heng Zhang
Summary: With the development of unmanned vehicles and other technologies, the demand for scene semantic segmentation is increasing. To improve the segmentation accuracy and efficiency, a detail guided multilateral segmentation network is proposed, which utilizes a multipath structure to process underlying and semantic information, and includes a feature fusion module to fuse semantic and detail information.
APPLIED SCIENCES-BASEL
(2022)
Article
Computer Science, Software Engineering
Jiayi Shi, Taiyong Li, Jiaxuan Xu
Summary: This paper proposes a recursive lightweight CNN approach (PPNets) that achieves significant improvement in image denoising, with fewer model parameters compared to traditional models and state-of-the-art CNN models.
Article
Physics, Multidisciplinary
Qi Li, Yunqing Liu, Quanyang Liu, Qiong Zhang, Fei Yan, Yimin Ma, Xinyu Zhang
Summary: This paper proposes a novel EEG emotion recognition method that effectively retains spatial and temporal information of EEG signals. The method utilizes a multidimensional feature structure, depthwise separable convolution, and ordered neuronal long short-term memory network to improve emotion classification performance.
Article
Biotechnology & Applied Microbiology
Juntong Yun, Du Jiang, Ying Liu, Ying Sun, Bo Tao, Jianyi Kong, Jinrong Tian, Xiliang Tong, Manman Xu, Zifan Fang
Summary: This article proposes a real-time target detection method based on a lightweight convolutional neural network, improving target detection technology by reducing the number of model parameters and improving detection speed. Experimental results demonstrate the effectiveness and superiority of the proposed method in complex scenes, with tests on video and deployment on the Android platform also confirming its real-time performance and scalability.
FRONTIERS IN BIOENGINEERING AND BIOTECHNOLOGY
(2022)
Article
Computer Science, Information Systems
Wenhao Yu, Guanwen Wang
Summary: This study proposes a novel ensemble recognition method that represents trajectory data using a graph structure and achieves good results in experiments. This method can help to discover the movement patterns of urban residents and provide effective assistance for city management.
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE
(2023)
Correction
Chemistry, Analytical
Tao Ye, Baocheng Wang, Ping Song, Juan Li
Article
Engineering, Civil
Tao Ye, Xi Zhang, Yi Zhang, Jie Liu
Summary: The proposed object-detection method using DFF-Net showed significant higher performance and was more effective and efficient for object detection in railway tracks compared to current state-of-the-art detectors. Evaluation results based on PASCAL VOC2007 and VOC2012 also indicated that the proposed method outperformed the state-of-the-art methods.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2021)
Article
Automation & Control Systems
Tao Ye, Cong Ren, Xi Zhang, Guodong Zhai, Rui Wang
Summary: This article proposes an object detector for collision warning in a train safety system, which has low power consumption and robust detection capability. By combining stable sampling, lightweight feature extraction, and feature fusion modules, the detection accuracy of multiscale and small objects is improved.
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
(2021)
Article
Engineering, Civil
Tao Ye, Jun Zhang, Zongyang Zhao, Fuqiang Zhou
Summary: Detection of railway traffic objects is crucial for safe train driving. This study proposes a novel deep learning method, MMFE-Net, to accurately detect railway objects. The network utilizes improved backbone network, spatial feature extraction, and attention fusion enhance module to address challenges in complex railway scenes. Experimental results show that MMFE-Net outperforms other methods on railway traffic dataset and is feasible for practical railway object detection tasks.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Engineering, Civil
Tao Ye, Zongyang Zhao, Shouan Wang, Fuqiang Zhou, Xiaozhi Gao
Summary: The article introduces a method that provides stable lightweight feature extraction and adaptive feature fusion network for real-time detection of obstacles in railway traffic scenarios. The method improves multiscale detection accuracy in complex environments and achieves better results than the state-of-the-art models in experiments.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Engineering, Electrical & Electronic
Tao Ye, Wenyang Qin, Zongyang Zhao, Xiaozhi Gao, Xiangpeng Deng, Yu Ouyang
Summary: Unmanned aerial vehicles (UAVs) are crucial for conducting automatic patrol inspections in cities, ensuring the safety of residents and normal city operation. However, issues such as difficulty in detecting small objects and severe occlusion in UAV images, as well as real-time performance requirements, arise during inspections. To address these, a real-time object detection network, feature fusion module, lightweight feature extraction module, and efficient convolutional transformer block-based convolutional multihead self-attention are proposed. These methods improve detection accuracy and speed in UAV image datasets and embedded devices.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Engineering, Electrical & Electronic
Yuchan Jie, Xiaosong Li, Fuqiang Zhou, Tao Ye
Summary: This article presents a tri-modal medical image fusion and denoising method, which has been evaluated through extensive experiments and outperformed the state-of-the-art methods in noise-free and noise fusions.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Engineering, Electrical & Electronic
Tao Ye, Xiangming Yan, Shouan Wang, Yunwang Li, Fuqiang Zhou
Summary: This article proposes an efficient approach for 3D point cloud place recognition based on feature point extraction and transformer (FPET-Net). By using a feature point extraction module and a point transformer module, the detection effect of place recognition can be improved and the model computation can be reduced. Experiments show significant improvements in parameter size and computation speed, and the method achieves excellent results in place recognition experiments.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)
Article
Engineering, Electrical & Electronic
Tao Ye, Wenyang Qin, Yunwang Li, Shouan Wang, Jun Zhang, Zongyang Zhao
Summary: This study proposes a global-local feature enhanced network (GLF-Net) for object detection in UAV images. The network utilizes a backbone and a multiscale feature fusion module to extract effective features, and a rotated regional proposal network to accurately detect rotated objects. Experimental results demonstrate that the method achieves state-of-the-art detection accuracy on multiple datasets.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)
Article
Engineering, Electrical & Electronic
Tao Ye, Jun Zhang, Yunwang Li, Xi Zhang, Zongyang Zhao, Zezhong Li
Summary: In this article, a deep learning method called CT-Net is proposed for low-altitude small-object detection. It addresses the limitations of existing detection methods in accuracy, model size, and speed through the introduction of an attention-enhanced transformer block, a lightweight bottleneck module, and a directional feature fusion structure. Experimental results show that CT-Net outperforms other detectors on low-altitude small-object datasets and MS COCO.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)
Article
Computer Science, Information Systems
Tao Ye, Zhihao Zhang, Xi Zhang, Fuqiang Zhou