Article
Computer Science, Artificial Intelligence
Ladislav Lenc, Jiri Martinek, Pavel Kral, Anguelos Nicolao, Vincent Christlein
Summary: This paper describes a complex and flexible web framework for historical document manipulation and analysis with a focus on OCR. The framework contains eight modules to facilitate three main tasks. Experimental evaluation shows that the system is efficient and can save human labor.
Article
Neurosciences
Geng Du-Yan, Wang Jia-Xing, Wang Yan, Liu Xuan-Yu
Summary: This study utilizes heart-rate variability (HRV) signal to construct three automatic sleep staging models and compare their performance. The results show that convolutional neural network (CNN) achieves the best classification effect.
NEUROSCIENCE LETTERS
(2022)
Article
Chemistry, Multidisciplinary
Zohreh Khosrobeigi, Hadi Veisi, Ehsan Hoseinzade, Hanieh Shabanian
Summary: Optical Character Recognition (OCR) is used to convert images into editable text in various languages. The unique challenges in designing a Persian OCR system include continuity between characters, semicircles, dots, oblique lines, and left-to-right characters. The proposed framework, Bina, addresses these challenges by using Convolution Neural Network (CNN) and deep bidirectional Long-Short Term Memory (BLSTM). The results show that Bina achieves about 96% accuracy in Persian texts and 88% accuracy in Persian and English texts, outperforming the baseline algorithm.
APPLIED SCIENCES-BASEL
(2022)
Article
Multidisciplinary Sciences
Yuguang Chen, Jintao Huang, Hongbin Xu, Jincheng Guo, Linyong Su
Summary: A dynamic spatiotemporal graph attention network traffic flow prediction model based on the attention mechanism was proposed to improve the accuracy of traffic flow prediction under the influence of nearby time traffic flow disturbance. Experimental results on public data sets demonstrate the superiority of the proposed model compared to six baseline models.
SCIENTIFIC REPORTS
(2023)
Article
Computer Science, Artificial Intelligence
Tingke Wen, Yuanxing Xiao, Anqi Wang, Haizhou Wang
Summary: The development of blockchain technology has brought prosperity to the cryptocurrency market but has also made the blockchain platform a hotbed of crimes. This paper proposes a model based on a hybrid deep neural network to detect phishing scam accounts and verifies its effectiveness on Ethereum. The LBPS model performs better than existing methods and baseline models with an F1-score of 97.86%.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Debashis Das Chakladar, Pradeep Kumar, Partha Pratim Roy, Debi Prosad Dogra, Erik Scheme, Victor Chang
Summary: The proposed multi-modal Siamese Neural Network (mSNN) combines EEG signals and offline signatures for improved user verification accuracy. The model achieved a classification accuracy of 98.57% on a dataset of 70 users, outperforming the current state-of-the-art by 12.86%.
INFORMATION FUSION
(2021)
Article
Computer Science, Artificial Intelligence
Gang Lv, Yining Sun, Fudong Nian, Maofei Zhu, Wenliang Tang, Zhenzhen Hu
Summary: This paper proposes a novel model called CLIP-OCR and Master Object (dubbed as COME) for text image captioning. It enhances the multimodal representation of OCR tokens and purifies the OCR-oriented scene graph with the concept of master object. The effectiveness of the proposed framework is demonstrated through experiments.
IMAGE AND VISION COMPUTING
(2023)
Article
Chemistry, Analytical
Ke-Wei Chen, Laura Bear, Che-Wei Lin
Summary: This study improves on the current solution methods for electrocardiographic imaging (ECGi) using machine learning and deep learning frameworks. By simultaneously recording electrocardiograms from pigs' ventricles and body surfaces, and constructing a model using neural networks, the study finds that relatively small datasets can achieve accuracy compatible with current standard methods.
Article
Computer Science, Information Systems
Jose Carlos Aradillas, Juan Jose Murillo-Fuentes, Pablo M. Olmos
Summary: In this paper, we successfully reduced the error rate in offline handwritten text recognition in historical documents by analyzing transfer learning, data augmentation, and error labeling algorithms, especially effective for training sets with few samples and errors.
Article
Engineering, Civil
Daniele Secci, Maria Giovanna Tanda, Marco D'Oria, Valeria Todaro
Summary: This study develops three different artificial intelligence models to investigate the effects of climate change on groundwater resources. Among them, the Long-Short Term Memory Neural Network (LSTM) performs the best in validation and testing phases, providing accurate predictions for the future.
JOURNAL OF HYDROLOGY
(2023)
Article
Geochemistry & Geophysics
Hongqi Zhang, Xudong Sun, Yuan Zhu, Fengqiang Xu, Xianping Fu
Summary: This paper proposes a novel global-local spectral weight network based on attention (GLSWA) for band selection, which combines fully connected neural network (FCN) and convolutional neural network (CNN). By designing attention-based scoring module (ASM) and convolutional reconstruction module (CRM), the attention of each band is adjusted by considering both the entire band features and the successive ones, achieving satisfactory accuracy.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Correction
Computer Science, Artificial Intelligence
Jiri Martinek, Ladislav Lenc, Pavel Kral
Summary: With the author(s) choosing Open Choice, the copyright of the article was changed on December 3, 2020 to [The Authors] [2020], and the article was immediately distributed under the terms of copyright.
NEURAL COMPUTING & APPLICATIONS
(2022)
Article
Computer Science, Software Engineering
Ishak Dolek, Atakan Kurt
Summary: This article presents a deep learning model and an OCR tool for printed Ottoman documents, which outperforms other OCR tools or models in terms of character recognition and word recognition accuracy. The hybrid model shows significant improvement in both character and word recognition rates.
CONCURRENCY AND COMPUTATION-PRACTICE & EXPERIENCE
(2022)
Article
Computer Science, Information Systems
Yingfeng Cai, Ruidong Zhao, Hai Wang, Long Chen, Yubo Lian, Yilin Zhong
Summary: This paper aims to establish a highly accurate, fast, and generalizable driving style recognition method. It addresses the lack of data types in driving style classification task and the low recognition accuracy of widely used unsupervised clustering algorithms and single convolutional neural network methods. The proposed method collects driver's operation time sequence information and extracts driver's style features through convolutional neural network. The Long Short Term Memory networks (LSTM) module is added to encode and transform the driving features for driving style classification. Results show over 93% accuracy and significantly improved speed in driving style recognition.
Article
Construction & Building Technology
Krisada Chaiyasarn, Apichat Buatik, Hisham Mohamad, Mingliang Zhou, Sirisilp Kongsilp, Nakhorn Poovarodom
Summary: This paper proposes an advanced inspection reporting system based on an integrated CNN-FCN crack detection system, which enables crack inspection and display for larger structures. The system utilizes a trained CNN to detect crack patches and a trained FCN system to segment cracks at the pixel-level. The system shows promising results with high accuracy and precision.
AUTOMATION IN CONSTRUCTION
(2022)
Correction
Computer Science, Artificial Intelligence
Jiri Martinek, Ladislav Lenc, Pavel Kral
Summary: With the author(s) choosing Open Choice, the copyright of the article was changed on December 3, 2020 to [The Authors] [2020], and the article was immediately distributed under the terms of copyright.
NEURAL COMPUTING & APPLICATIONS
(2022)
Article
Computer Science, Interdisciplinary Applications
Dalibor Fiala, Pavel Kral, Martin Dostal
Summary: The study tested the hypothesis that computer science papers with questions in their titles are cited more frequently. The analysis of data from almost two million computer science papers showed that papers with questions receive an average of 20% more citations than other papers, which is statistically significant.
Article
Computer Science, Artificial Intelligence
Lysimachos Maltoudoglou, Andreas Paisios, Ladislav Lenc, Jiri Martinek, Pavel Kral, Harris Papadopoulos
Summary: In this study, the researchers present a novel approach to improve the computational efficiency of Label Powerset Inductive Conformal Prediction in multi-label text classification. Experimental results show that contextualised-based classifiers outperform non-contextualised ones and achieve state-of-the-art performance across all datasets.
PATTERN RECOGNITION
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Ladislav Lenc, Jiri Martinek, Pavel Kral
Summary: This paper presents a novel approach to page segmentation into text lines, which is used as input for a line-based OCR system. The approach decomposes the problem into text-block and text-line segmentation and employs algorithms based on fully convolutional neural networks. The proposed method is evaluated on standard corpora and a new dataset created from freely available data.
ARTIFICIAL INTELLIGENCE AND SOFT COMPUTING, ICAISC 2022, PT II
(2023)
Proceedings Paper
Acoustics
Jiri Martinek, Christophe Cerisara, Pavel Kral, Ladislav Lenc, Josef Baloun
Summary: Large pre-trained language models have achieved impressive results in zero-shot learning, but it is still challenging to design effective prompts for certain tasks like dialogue act recognition. We propose an alternative approach that replaces manual prompts with simple rules, which are more intuitive and easier to design. Our experiments on question type recognition demonstrate that this approach can achieve competitive performances and we analyze its limitations.
Proceedings Paper
Computer Science, Artificial Intelligence
Ladislav Lenc, Jiri Martinek, Josef Baloun, Martin Prantl, Pavel Kral
Summary: The paper introduces a method for the detection, classification, and recognition of toponyms in hand-drawn historical cadastral maps. The detected and recognized toponyms are used as keywords for intelligent and efficient searching in historical map collections. The paper proposes a novel approach for toponym classification based on the KAZE descriptor and evaluates several state-of-the-art methods for text and object detection on the toponym detection task. Additionally, the paper presents the results of toponym text recognition using the popular Tesseract engine.
DOCUMENT ANALYSIS SYSTEMS, DAS 2022
(2022)
Proceedings Paper
Computer Science, Information Systems
Jiri Martinek, Pavel Kral, Ladislav Lenc
Summary: This paper focuses on dialogue act recognition from printed documents and introduces a novel deep model for visual DA recognition. The study shows that visual information does not impact DA recognition on high-quality images, but significantly improves the score on low-quality images with erroneous OCR. This is the first attempt to focus on DA recognition from visual data.
DOCUMENT ANALYSIS AND RECOGNITION - ICDAR 2021, PT II
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Joseph Chazalon, Edwin Carlinet, Yizi Chen, Julien Perret, Bertrand Dumenieu, Clement Mallet, Thierry Geraud, Vincent Nguyen, Nam Nguyen, Josef Baloun, Ladislav Lenc, Pavel Kral
Summary: This paper presents the final results of the ICDAR 2021 MapSeg competition, which focuses on historical map segmentation of a series of historical atlases of Paris, France. The winning teams used different network structures and methods for each task. The research outcomes have a positive impact on the development of historical map segmentation technology.
DOCUMENT ANALYSIS AND RECOGNITION, ICDAR 2021, PT IV
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Josef Baloun, Pavel Kral, Ladislav Lenc
Summary: This research focuses on the segmentation of historical handwritten documents, specifically chronicles, using a fully convolutional neural network approach. A new dataset was created, consisting of 58 images with precise annotations for text, image, and graphic regions at a pixel level. Multiple experiments were conducted to identify the best method configuration, including a novel data augmentation method.
ICAART: PROCEEDINGS OF THE 13TH INTERNATIONAL CONFERENCE ON AGENTS AND ARTIFICIAL INTELLIGENCE - VOL 2
(2021)