Article
Computer Science, Artificial Intelligence
Yong Soon Tan, Kian Ming Lim, Chin Poo Lee
Summary: Hand gesture recognition is essential for human communication and interaction, and has applications in human-computer interaction and bridging language barriers. Hand-crafted and deep learning approaches can tackle challenges in vision-based hand gesture recognition, with the former focusing on specific challenges and the latter adapting to various challenges through supervised learning.
EXPERT SYSTEMS WITH APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Guoqing Li, Meng Zhang, Jiaojie Li, Feng Lv, Guodong Tong
Summary: The paper introduces two novel and efficient lightweight CNN architectures, DenseDsc and Dense2Net, which improve parameter efficiency by using dense connectivity and different convolution methods.
PATTERN RECOGNITION
(2021)
Article
Computer Science, Information Systems
Yanhui Xiao, Huawei Tian, Gang Cao, Duo Yang, Hui Li
Summary: This paper proposes a PRNU extraction algorithm based on DHDN, which can better handle natural noise and improve the accuracy of source camera identification.
MULTIMEDIA TOOLS AND APPLICATIONS
(2022)
Article
Biochemistry & Molecular Biology
Xiaobing Zhang, Yin Hu, Wen Chen, Gang Huang, Shengdong Nie
Summary: A novel segmentation method integrating three densely connected 2D convolutional neural networks was proposed to address the computational burden of processing 3D medical scans and the lack of spatial information in 2D medical scans. Through network design modifications and improved loss functions, the method demonstrated promising accuracy and fast processing in medical image segmentation.
JOURNAL OF ZHEJIANG UNIVERSITY-SCIENCE B
(2021)
Article
Environmental Sciences
Feng Zhao, Junjie Zhang, Zhe Meng, Hanqiang Liu
Summary: A densely connected pyramidal dilated convolutional network (PDCNet) is proposed in this paper to address the issue of blind spots in the receptive field caused by dilated convolution, achieving continuous and multi-scale receptive fields. By utilizing a pyramid pattern of dilated convolutional layers and a feature fusion mechanism, the network demonstrates good classification performance in HSI compared to other popular models.
Article
Geochemistry & Geophysics
Jinglu He, Yinghua Wang, Hongwei Liu
Summary: This article proposes a method for more effective ship classification in MR SAR images by extending dense convolutional networks and utilizing a multitask learning framework, joint minimizing softmax log-loss and triplet loss for better deep feature extraction. Experiments on a MR SAR ship data set show superior performance compared to CNN models for 3- and 5-class recognition tasks.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2021)
Article
Computer Science, Artificial Intelligence
Yaoming Cai, Zijia Zhang, Qin Yan, Dongfang Zhang, Mst Jainab Banu
Summary: The proposed Dense Connected Convolutional ELM (DC2ELM) is a simple yet effective deep ELM method for spectral-spatial classification of hyperspectral images. By introducing dense connections and stacked ELM auto-encoders, it can make full use of intermediate feature maps and achieve a deeper architecture, with fewer trainable parameters compared to traditional convolutional neural networks.
Article
Automation & Control Systems
Shuqiang Wang, Xiangyu Wang, Yanyan Shen, Bing He, Xinyan Zhao, Prudence Wing-Hang Cheung, Jason Pui Yin Cheung, Keith Dip-Kei Luk, Yong Hu
Summary: Assessment of skeletal maturity is crucial for clinicians to make treatment decisions, but using machine learning for this task is challenging. In this article, an ensemble-based deep learning approach is proposed to automatically assess the maturity of the radius and ulna from left-hand X-ray images. Experimental results demonstrate the effectiveness of the proposed model.
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS
(2022)
Article
Chemistry, Analytical
Zhuo Chen, Dazhi Gao, Kai Sun, Xiaojing Zhao, Yueqi Yu, Zhennan Wang
Summary: In indoor environments, reverberation poses a challenge to sound classification. To overcome this, we used a DenseNet to combine speech spectral features and achieved a classification accuracy of 95.90%, better than other CNNs. The optimized DenseNet model size is only 3.09 MB and can be deployed on embedded devices.
Article
Biochemistry & Molecular Biology
Mobeen Ur Rehman, Hilal Tayara, Kil To Chong
Summary: The study presents a neural network-based tool, DCNN-4mC, for identifying DNA N4-methylcytosine (4mC) sites. By combining all available datasets of different species to create a single benchmark dataset for each species, the tool's performance was evaluated on 12 different species. DCNN-4mC achieved higher accuracy compared to state-of-the-art tools on various datasets of different species and demonstrated high performance on independent test datasets as well.
COMPUTATIONAL AND STRUCTURAL BIOTECHNOLOGY JOURNAL
(2021)
Article
Biochemical Research Methods
Pi-Jing Wei, Zhen-Zhen Pang, Lin-Jie Jiang, Da-Yu Tan, Yan-Sen Su, Chun-Hou Zheng
Summary: This study proposes a method called DNPPro based on densely connected convolutional neural networks to predict promoters in Nannochloropsis. The method collects promoter sequences, removes similarity, and constructs a robust classifier. Experimental results verify the generalization ability of the method and its effectiveness in distinguishing promoters.
Article
Computer Science, Information Systems
Sunny Dayal Vanambathina, Manaswini Burra, Bhumika Edupalli, Eswar Reddy Vallem, Venkata Sravani Nellore
Summary: This paper presents a fully convolutional neural network for enhancing real-time speech in the time domain. It includes skip connections and dilated convolutions to aggregate contextual information. The utilization of causal convolutions prevents information inflow from subsequent frames, making it ideal for real-time applications.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Engineering, Electrical & Electronic
Jie Xie, Nanjun He, Leyuan Fang, Pedram Ghamisi
Summary: The paper introduces a multiscale densely-connected convolutional network (MS-DenseNet) framework that efficiently utilizes multi-scale information for hyperspectral image classification. Experimental results demonstrate the superiority of the proposed MS-DenseNet over single scale-based CNN classification model and other well-known classification methods.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2021)
Article
Computer Science, Information Systems
Gurinder Singh, Puneet Goyal
Summary: The article introduces a new general-purpose forensic approach based on a shallow densely connected convolutional neural network, which outperforms existing state-of-the-art general-purpose forensic schemes in multiple image manipulation detection.
ACM TRANSACTIONS ON MULTIMEDIA COMPUTING COMMUNICATIONS AND APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Chenquan Gan, Junhao Xiao, Zhangyi Wang, Zufan Zhang, Qingyi Zhu
Summary: This paper proposes a novel method for facial expression recognition by eliminating redundant information from emotional-unrelated regions, leading to more accurate recognition. The method utilizes a densely connected convolutional neural network with hierarchical spatial attention, and its superior performance is verified through experiments.
IMAGE AND VISION COMPUTING
(2022)