Article
Computer Science, Artificial Intelligence
Xiangli Yang, Zixing Song, Irwin King, Zenglin Xu
Summary: This paper provides a comprehensive survey on deep semi-supervised learning methods, including model design and unsupervised loss functions. It categorizes existing methods into different types and reviews 60 representative methods with a detailed comparison. The paper also discusses the shortcomings of existing methods and proposes heuristic solutions.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Environmental Sciences
Brandon Hobley, Riccardo Arosio, Geoffrey French, Julie Bremner, Tony Dolphin, Michal Mackiewicz
Summary: Intertidal seagrass plays a vital role in coastal environments, but most habitats have been declining due to human impacts. This study compares two methods for classifying seagrass habitats- OBIA and FCNNs. Using semi-supervision, we demonstrate the utility of FCNNs in mapping seagrass and other coastal features.
Article
Biology
Wenxue Li, Wei Lu, Jinghui Chu, Qi Tian, Fugui Fan
Summary: In this paper, a novel Confidence-Guided Mask Learning (CGML) method is proposed for semi-supervised medical image segmentation. The method introduces an auxiliary generation task with mask learning and a confidence-guided masking strategy to improve segmentation accuracy and feature representation reliability.
COMPUTERS IN BIOLOGY AND MEDICINE
(2023)
Article
Computer Science, Software Engineering
Zhenyu Shu, Sipeng Yang, Haoyu Wu, Shiqing Xin, Chaoyi Pang, Ladislav Kavan, Ligang Liu
Summary: This paper proposes a novel semi-supervised algorithm for 3D shape segmentation, which achieves satisfactory results by locating seed faces with simple interaction and automatically learning label information.
COMPUTER-AIDED DESIGN
(2022)
Article
Computer Science, Artificial Intelligence
Krishna Chaitanya, Ertunc Erdil, Neerav Karani, Ender Konukoglu
Summary: Supervised deep learning methods require large labeled datasets for accurate medical image segmentation. This paper proposes a local contrastive loss-based approach that utilizes pseudo-labels of unlabeled images and limited annotated images to learn pixel-level features for segmentation. Experimental results on three public medical datasets demonstrate the substantial improvement achieved by the proposed method.
MEDICAL IMAGE ANALYSIS
(2023)
Article
Chemistry, Analytical
Luca Caltagirone, Mauro Bellone, Lennart Svensson, Mattias Wahde, Raivo Sell
Summary: This study investigates the improvement of semantic segmentation performance by fusing lidar and camera data in a supervised learning context, as well as the impact of fusion on semi-supervised learning. The results show that semi-supervised learning and fusion techniques increase the network's overall performance in challenging scenarios.
Article
Neurosciences
Gaoxiang Chen, Jintao Ru, Yilin Zhou, Islem Rekik, Zhifang Pan, Xiaoming Liu, Yezhi Lin, Beichen Lu, Jialin Shi
Summary: A novel semi-supervised segmentation framework integrating improved mean teacher and adversarial network was proposed, which includes multi-scale feature consistency loss and shape-aware embedding scheme. The experiments demonstrated that the method can effectively leverage unlabeled data and outperform other semi-supervised methods trained with the same labeled data.
Article
Computer Science, Artificial Intelligence
Krishna Chaitanya, Neerav Karani, Christian F. Baumgartner, Ertunc Erdil, Anton Becker, Olivio Donati, Ender Konukoglu
Summary: Supervised learning-based segmentation methods typically require a large number of annotated training data, which is challenging in medical applications. This work presents a novel task-driven data augmentation method that significantly outperforms other approaches in limited annotation settings.
MEDICAL IMAGE ANALYSIS
(2021)
Article
Biology
Shajahan Aboobacker, Deepu Vijayasenan, Sumam David, Pooja K. Suresh, Saraswathy Sreeram
Summary: This study aims to predict the malignancy in effusion cytology images and reduce scanning time using deep learning models. The authors extend two semi-supervised learning models and introduce reverse augmentation to address spatial alterations in image annotation. The results show that the extended models improve the accuracy of malignancy pixel prediction and successfully save scanning time compared to the baseline model.
COMPUTERS IN BIOLOGY AND MEDICINE
(2022)
Article
Computer Science, Artificial Intelligence
Anneke Meyer, Suhita Ghosh, Daniel Schindele, Martin Schostak, Sebastian Stober, Christian Hansen, Marko Rak
Summary: The study introduces a semi-supervised learning technique named uncertainty-aware temporal self-learning (UATS) for fine-grained segmentation of the prostate, which significantly outperforms the supervised baseline in segmentation quality, especially for minimal amounts of labeled data.
ARTIFICIAL INTELLIGENCE IN MEDICINE
(2021)
Article
Computer Science, Artificial Intelligence
Junwen Pan, Pengfei Zhu, Kaihua Zhang, Bing Cao, Yu Wang, Dingwen Zhang, Junwei Han, Qinghua Hu
Summary: This paper introduces a Self-supervised Low-Rank Network (SLRNet) for semantic segmentation with limited annotations, which improves the robustness and generalization performance of the model by learning precise pseudo-labels through cross-view self-supervision and collective matrix factorization.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2022)
Article
Biochemistry & Molecular Biology
Qiang Lin, Runxia Gao, Mingyang Luo, Haijun Wang, Yongchun Cao, Zhengxing Man, Rong Wang
Summary: This study proposes a semi-supervised segmentation model for automated identification and delineation of skeletal metastasis lesions in bone scan images, aiding in clinical diagnosis.
FRONTIERS IN MOLECULAR BIOSCIENCES
(2022)
Article
Construction & Building Technology
Ankang Ji, Yunxiang Zhou, Limao Zhang, Robert L. K. Tiong, Xiaolong Xue
Summary: This research proposes a deep learning method called SPCNet to boost multi-class tunnel point cloud segmentation by alleviating labeling tasks. The results show that SPCNet outperforms other state-of-the-art methods and has great potential for practical applications.
AUTOMATION IN CONSTRUCTION
(2023)
Article
Chemistry, Analytical
Yusen Wan, Liang Gao, Xinyu Li, Yiping Gao
Summary: Printed circuit board (PCB) defect detection is crucial in PCB production. This paper proposes a semi-supervised defect detection method that leverages unlabeled samples and introduces data-expanding and batch-adding strategies to achieve competitive results with fewer labeled samples.
Article
Computer Science, Software Engineering
Guanhua Li, Shuang Yang, Siming Cao, Weidong Zhu, Yinglin Ke
Summary: This paper introduces the use of semi-supervised learning and an improved texture segmentation algorithm for accurate detection of drilled holes on composite parts. By training a superior circle segmentation model with partially annotated data, massive data labeling can be avoided. Experimental results demonstrate the effectiveness of the method for detecting circular holes with different texture information commonly found in robotic drilling, achieving a measurement accuracy of 0.03 mm.