Article
Chemistry, Multidisciplinary
Berardino Prencipe, Nicola Altini, Giacomo Donato Cascarano, Antonio Brunetti, Andrea Guerriero, Vitoantonio Bevilacqua
Summary: Liver segmentation is a crucial step in surgical planning from computed tomography scans. The researchers in this study focused on the liver segments' classification task and used a Convolutional Neural Network (CNN) trained with a specific loss function to achieve impressive results.
APPLIED SCIENCES-BASEL
(2022)
Article
Engineering, Biomedical
Michael Yeung, Evis Sala, Carola-Bibiane Schoenlieb, Leonardo Rundo
Summary: Automatic segmentation methods using deep neural networks have advanced medical image analysis. However, class imbalance in medical datasets poses a challenge for model convergence. In this study, we propose the Unified Focal loss function to handle class imbalance and demonstrate its superior performance compared to other loss functions on various medical imaging datasets.
COMPUTERIZED MEDICAL IMAGING AND GRAPHICS
(2022)
Article
Chemistry, Multidisciplinary
Haotian Li, Zhuang Yue, Jingyu Liu, Yi Wang, Huaiyu Cai, Kerang Cui, Xiaodong Chen
Summary: This paper proposed a pixel-level crack segmentation network called SCCDNet based on convolutional neural networks, which achieved the best crack segmentation performance with an F-score of 0.7763 by using techniques such as depthwise separable convolution. The network was trained and tested on a dataset containing cracks collected in different scenes, demonstrating its effectiveness in detecting cracks accurately and efficiently.
APPLIED SCIENCES-BASEL
(2021)
Article
Environmental Sciences
Foivos I. Diakogiannis, Francois Waldner, Peter Caccetta
Summary: This study presents a deep learning framework for semantic change detection in high-resolution aerial images, featuring new building blocks, loss function, attention module, and backbone architecture tailored for this task. The key insight is to facilitate relative attention fusion between two convolution layers.
Article
Engineering, Civil
Quang Du Nguyen, Huu-Tai Thai
Summary: This paper conducts a large-scale performance comparison of twelve commonly used loss functions on four benchmark datasets, considering accuracy, sensitivity to threshold change, and varying degrees of imbalance severity. The results show that weighted binary cross-entropy loss, Focal loss, Dice-based loss, and compound loss functions outperform others as imbalance severity increases.
ENGINEERING STRUCTURES
(2023)
Article
Computer Science, Artificial Intelligence
Yimin Ou, Rui Yang, Lufan Ma, Yong Liu, Jiangpeng Yan, Shang Xu, Chengjie Wang, Xiu Li
Summary: Proposed a box-free and NMS-free end-to-end instance segmentation framework, UniInst, which generates a unique representation for each instance through an instance-aware one-to-one assignment scheme and integrates a novel prediction re-ranking strategy for improved discriminative learning.
Article
Chemistry, Multidisciplinary
Xiaohu Zhang, Haifeng Huang
Summary: Given the significance of convolutional neural network-based attention models in road maintenance, a PSNet method with a Parallel Convolution Module (PCM) and Self-Gated Attention Block (SGAB) was proposed. Experimental results demonstrated competitive segmentation performance for crack detection, showing significant improvements over traditional attention models.
APPLIED SCIENCES-BASEL
(2023)
Article
Computer Science, Artificial Intelligence
Nhung Hong Thi Nguyen, Stuart Perry, Don Bone, Ha Thanh Le, Thuy Thi Nguyen
Summary: This paper proposes a new method utilizing a two-stage convolutional neural network for road crack detection and segmentation, which outperforms existing approaches in handling noisy, low-resolution images, and imbalanced datasets with an F1-measure of over 0.91 on three datasets.
EXPERT SYSTEMS WITH APPLICATIONS
(2021)
Article
Engineering, Electrical & Electronic
Jacob Konig, Mark David Jenkins, Mike Mannion, Peter Barrie, Gordon Morison
Summary: This work proposes optimized deep encoder-decoder methods for surface crack segmentation, which combine various techniques to enhance performance. Different encoder strategies and a data augmentation policy are studied to achieve improved results. The introduction of two new techniques, such as test-time augmentation and statistical result analysis, leads to state-of-the-art performance in all datasets.
DIGITAL SIGNAL PROCESSING
(2021)
Article
Engineering, Biomedical
Chengjian Qiu, Zhe Liu, Yuqing Song, Jing Yin, Kai Han, Yan Zhu, Yi Liu, Victor S. Sheng
Summary: Accurate pancreas segmentation is crucial for pancreatic cancer diagnosis. We propose RTUNet, a residual transformer UNet, to overcome the challenges of position changes, shape variability, and blurred boundary in pancreas segmentation. Our method achieves superior performance compared to the state-of-the-art on the NIH dataset, with an 86.25% Dice similarity coefficient (DSC). Ablation studies confirm the effectiveness of each proposed module.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2023)
Article
Computer Science, Software Engineering
Dan Wang, Guoqing Hu, Chengzhi Lyu
Summary: The paper introduces a lightweight end-to-end feature refinement network (FRNet) to address the issue of spatial information loss in medical image segmentation. By incorporating spatial refinement path, semantic refinement path, and feature adaptive fusion block (FAF block), high accuracy is achieved in different tasks without the need for post-processing.
Article
Mechanics
Ce Xiao, Jean-Yves Buffiere
Summary: In this study, an image segmentation method based on convolutional neural network is developed to successfully extract the 3D shapes of internal fatigue cracks in metals, combined with a 'Hessian matrix' filter.
ENGINEERING FRACTURE MECHANICS
(2021)
Article
Computer Science, Artificial Intelligence
Yao Chen, Yanan Sun, Jiancheng Lv, Bijue Jia, Xiaoming Huang
Summary: This paper introduces a novel method for heart sound segmentation based on convolutional long short-term memory (CLSTM) which directly uses audio recording as input, improving robustness and adaptability in processing HSS tasks. The algorithm does not require feature extraction in advance and demonstrates outstanding performance on real-world PCG datasets.
COMPLEX & INTELLIGENT SYSTEMS
(2021)
Review
Construction & Building Technology
Raza Ali, Joon Huang Chuah, Mohamad Sofian Abu Talip, Norrima Mokhtar, Muhammad Ali Shoaib
Summary: This article reviews the application of Convolutional Neural Networks (CNN) in civil structure crack detection, emphasizing the importance of CNN in image classification and segmentation, as well as analyzing recent developments. In addition to discussing machine learning methods, it also introduces the limitations of manual processing and image processing techniques in crack detection.
AUTOMATION IN CONSTRUCTION
(2022)
Article
Engineering, Civil
Keqiang Li, Hui Xiong, Dameng Yu, Jinxin Liu, Yu'ang Guo, Jianqiang Wang
Summary: This method utilizes an end-to-end neural network model for road detection in autonomous driving, taking advantage of road boundary characteristics and multi-task learning. By reassigning labels and rebalancing losses, it focuses on learning hard examples on the boundary to enhance performance. A data augmentation method based on road geometric transformation is proposed for network robustness.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)