Article
Computer Science, Artificial Intelligence
Pengfei Xian, Lai-Man Po, Jingjing Xiong, Chang Zhou, Yuzhi Zhao, Wing-Yin Yu, Weifeng Ou, Yujia Zhang, Xiaori Zhang
Summary: This study proposes a novel Pixel Voting Decoder to meet the performance requirements of both instance segmentation and semantic segmentation tasks. By regressing the interlayer pixel relationships between the input and output feature maps, the decoder dynamically decodes the higher level information from the encoder. The matrix computation for dynamic deconvolution enhances calculation efficiency.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Xinyu Zhou, Xuanya Li, Kai Hu, Yuan Zhang, Zhineng Chen, Xieping Gao
Summary: This study introduces an efficient 3D residual neural network for brain tumor segmentation, which has less computational complexity and GPU memory consumption. It utilizes a computation-efficient encoder and decoder structure, along with a fusion loss function to improve convergence and tackle data imbalance issues, resulting in excellent experimental results.
EXPERT SYSTEMS WITH APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Mohammad Golbabaee, Guido Buonincontri, Carolin M. Pirkl, Marion Menzel, Bjoern H. Menze, Mike Davies, Pedro A. Gomez
Summary: A novel pipeline for multi-parametric quantitative MRI image computing is proposed, utilizing compressed sensing reconstruction and deep learned quantitative inference. The approach effectively recovers accurate and consistent quantitative information through flexible generation of rich training samples in the trained model.
MEDICAL IMAGE ANALYSIS
(2021)
Article
Computer Science, Interdisciplinary Applications
Gorkem Can Ates, Recep M. Gorguluarslan
Summary: This study proposes a two-stage network model for topology optimization, which effectively reduces structural disconnections and pixel-wise errors, enhancing the predictive performance of DNNs. The optimized framework improves network prediction ability while significantly reducing compliance and volume fraction errors.
STRUCTURAL AND MULTIDISCIPLINARY OPTIMIZATION
(2021)
Article
Computer Science, Information Systems
Md. Shahir Zaoad, M. M. Rushadul Mannan, Angshu Bikash Mandol, Mostafizur Rahman, Md Adnanul Islam, Md. Mahbubur Rahman
Summary: Video captioning is an automated process that generates captions for videos by understanding their content. This research focuses on Bengali video captioning, which is an underexplored area compared to English video captioning. The study implements sequence-to-sequence models like LSTM, BiLSTM, and GRU combined with CNN models VGG-19, Inceptionv3, and ResNet50v2 to extract video frame features and generate textual descriptions. Attention mechanism is also incorporated for the first time in Bengali video captioning. A novel Bengali video captioning dataset is created from Microsoft Research Video Description Corpus (MSVD) dataset, and the model's performance is evaluated using popular metrics such as BLEU, METEOR, and ROUGE. The proposed attention-based hybrid model outperforms existing models and sets a new benchmark for Bengali video captioning.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2023)
Article
Environmental Sciences
Weisheng Li, Minghao Xiang, Xuesong Liang
Summary: A dense encoder-decoder network with feedback connections is proposed for pan-sharpening to meet the need for multispectral images with high spatial resolution. The network consists of four parts: feature extraction, multiscale feature-extraction, feature fusion and recovery, and continuous feedback connections. Experiments on various satellite datasets show significant improvements in spectral quality and spatial details compared to existing methods.
Article
Computer Science, Interdisciplinary Applications
Zhi-Xiong Lan, Xue-Mei Dong
Summary: With the advancement of deep learning, the proposed neural networks MiniCrack and MiniCrack-Light show superior performance in narrow crack detection compared to the current state-of-the-art networks, with fewer parameters and stronger robustness.
COMPUTERS IN INDUSTRY
(2022)
Article
Computer Science, Artificial Intelligence
Shirin Hajeb-M, Alicia Cascella, Matt Valentine, K. H. Chon
Summary: This study demonstrates a novel application of a deep convolutional neural network encoder-decoder method to suppress CPR artifact in near real-time using only ECG data. By training the network with CPR-contaminated ECG data, the proposed method enables continuous and accurate AED rhythm analysis without stopping CPR, resulting in improved survival rates for out-of-hospital cardiac arrest patients.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Shibai Yin, Yibin Wang, Yee-Hong Yang
Summary: A new attentive U-recurrent encoder-decoder dehazing network is proposed to address the three main limitations of convolutional neural networks in image dehazing, including ignoring relevant haze information, spatial inconsistency, and insufficient receptive field. The network combines an attentive recurrent network and a U-recurrent encoder-decoder network to improve spatial consistency, reduce information dilution, and enhance structural information capture. The experimental results demonstrate superior performance compared to state-of-the-art dehazing algorithms on both synthetic and real hazy images.
Article
Chemistry, Multidisciplinary
Jie Liu, Qiu Tang, Wei Qiu, Jun Ma, Yuhong Qin, Biao Sun
Summary: This paper introduces a novel diagnosis strategy for power quality based on unsupervised learning, using a residual denoising convolutional auto-encoder (RDCA) to automatically extract features from complex power quality disturbances (PQDs). Additionally, a classification framework combining RDCA and SCNN is proposed for classifying complex PQDs, with visual analysis used to better understand the model and features from different layers. Simulation and experimental tests validate the practicality and robustness of the RDCA.
APPLIED SCIENCES-BASEL
(2021)
Article
Computer Science, Artificial Intelligence
Huseyin Uzen, Muammer Turkoglu, Berrin Yanikoglu, Davut Hanbay
Summary: This study proposes a Swin transformer-based model called Multi-Feature Integration Network (Swin-MFINet) for pixel-level surface defect detection. The proposed model combines different feature extraction and fusion methods through encoder, decoder, and Multi-Feature Integration modules, achieving superior performance.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Plant Sciences
Helong Yu, Minghang Che, Han Yu, Yuntao Ma
Summary: This paper proposes a lightweight weed segmentation network model to improve the weed detection capability of mobile weed control devices. Experimental results show that the proposed model achieves high segmentation accuracy on a soybean field weed dataset.
FRONTIERS IN PLANT SCIENCE
(2023)
Article
Computer Science, Software Engineering
Pengfei Wang, Yunqi Li, Yaru Sun, Dongzhi He, Zhiqiang Wang
Summary: This paper presents a gastric cancer lesion dataset for gastric tumor image segmentation research and proposes a multiscale boundary neural network (MBNet) for automatically segmenting real tumor area in gastric cancer images. The experimental results demonstrate that the proposed method achieves high accuracy and similarity coefficient, outperforming existing approaches.
Review
Biology
Tao Zhou, QianRu Cheng, HuiLing Lu, Qi Li, XiangXiang Zhang, Shi Qiu
Summary: This paper reviews the image fusion methods based on deep learning from five aspects: the principle and advantages, the classification of methods, the application in medical image field, the evaluation metrics, and the challenges faced. It provides a systematic summary and guidance for the study of multi-modal medical images.
COMPUTERS IN BIOLOGY AND MEDICINE
(2023)
Article
Computer Science, Hardware & Architecture
Yujie Xu, Yongjun Zhang, Jingrong Cui, Zhongwei Cui, Yitong Yang
Summary: This work proposes a robust edge-guided and color-guided dehazing network (ECGDN) to address the issues of color distortion and blurred edge details in current dehazing methods. By using a cross-aggregation dual U-Net (CADU), a color feature extractor (CFE), a prior attention module (PAM), and a prior fusion group (PFG), ECGDN successfully mitigates these problems.
COMPUTERS & ELECTRICAL ENGINEERING
(2022)
Article
Computer Science, Interdisciplinary Applications
Prasun Chandra Tripathi, Soumen Bag
Summary: In this paper, a novel deep learning-based computer-aided diagnosis method for glioma tumors is proposed. The fusion of different residual networks enhances tumor classification performance, and the Dempster-Shafer Theory (DST)-based fusion technique produces superior results.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2022)
Article
Biochemical Research Methods
Prasun Chandra Tripathi, Soumen Bag
Summary: In this study, a Convolutional Neural Network (CNN)-based framework is proposed for non-invasive grading of tumors from 3D MRI scans. The framework includes two novel CNN architectures for tumor segmentation and glioma grading classification. Experimental results show that the proposed framework outperforms state-of-the-art methods.
IEEE-ACM TRANSACTIONS ON COMPUTATIONAL BIOLOGY AND BIOINFORMATICS
(2023)
Article
Computer Science, Artificial Intelligence
Prasun Chandra Tripathi, Soumen Bag
IET IMAGE PROCESSING
(2020)