检索结果-内蒙古大学图书馆

Real-time deep learning for multi-label retinal disease diagnosis with embedded system

学校读者我要写书评

暂无评论

Computational and Structural Biotechnology Reports

Computational and Structural Biotechnology Reports 2025年 2卷

作者： Feudjio Ghislain Saha Tchinda Beaudelaire Romain Atangana Tchiotsop Daniel Research Unit of Automation and Applied Computer (UR-AIA) Electrical Engineering Department of IUT-FV University of Dschang P.O. Box: 134 Bandjoun Cameroon Research Unit of Condensed Matter of Electronics and Signal Processing (UR‑MACETS) Department of Physics Faculty of Sciences University of Dschang P.O. Box 67 Dschang Cameroon Department of Computer Science Higher Teacher Training College University of Bertoua-Cameroon P.O.Box 652 Bertoua Cameroon

Background Ophthalmological consultations are essential for the early detection of retinal pathologies. The development of advanced diagnostic devices that combine portability, ease of access, and energy efficiency offers significant benefits, especially in rural or peri-urban areas, which often struggle with limited medical resources and time constraints. Method In this study, we develop a classification platform for various retinal diseases using a Raspberry Pi 4 board. The system relies on a lightweight deep learning model based on convolutional neural networks (CNNs). To enhance the model's efficiency, we integrate biorthogonal wavelet transforms, which facilitate the effective extraction of relevant features from the input images. This approach reduces computational complexity while maintaining the quality of the extracted information. The model, optimized for real-time deployment, was trained on a dataset containing 4303 retinal images, representing four classes of pathologies. Results The proposed embedded model achieves an accuracy of 0.9806 across the four classes of pathologies collected from the two public databases ODIR (Ocular Disease Intelligence Recognition) and RFMiD (Retinal Fundus Multi-Disease Image Dataset). Conclusions Compared to several state-of-the-art methods published to date, the outstanding performance of the proposed embedded system demonstrates its potential as a valuable tool for clinicians to diagnose various ocular diseases in underprivileged healthcare settings.

关键词： Embedded board Real-time classification Lightweight CNN model Wavelet transforms

Learning active spine behaviors for dynamic and efficient locomotion

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Bhattacharya, Shounak Singla, Abhik Abhimanyu Dholakiya, Dhaivat Bhatnagar, Shalabh Amrutur, Bharadwaj Ghosal, Ashitava Kolathaya, Shishir Robert Bosch Centre for Cyber-Physical Systems IISc Bangalore India Department of Mechanical Engineering BITS Pilani Goa campus India Faculty of Computer Science and Automation Faculty of Electrical and Computer Engineering Faculty of Mechanical Engineering

In this work, we provide a simulation framework to perform systematic studies on the effects of spinal joint compliance and actuation on bounding performance of a 16-DOF quadruped spined robot Stoch 2. Fast quadrupedal locomotion with active spine is an extremely hard problem, and involves a complex coordination between the various degrees of freedom. Therefore, past attempts at addressing this problem have not seen much success. Deep-Reinforcement Learning seems to be a promising approach, after its recent success in a variety of robot platforms, and the goal of this paper is to use this approach to realize the aforementioned behaviors. With this learning framework, the robot reached a bounding speed of 2:1 m/s with a maximum Froude number of 2. Simulation results also show that use of active spine, indeed, increased the stride length, improved the cost of transport, and also reduced the natural frequency to more realistic values. Copyright © 2019, The Authors. All rights reserved.

关键词： Deep learning

Pareto-wise Ranking Classifier for Multi-objective Evolutionary Neural Architecture Search

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Ma, Lianbo Li, Nan Yu, Guo Geng, Xiaoyu Cheng, Shi Wang, Xingwei Huang, Min Jin, Yaochu College of Software Northeastern University Shenyang110819 China The Institute of Intelligent Manufacturing Nanjing Tech University Nanjing211816 China The School of Computer Science Shaanxi Normal University Xi’an710062 China The College of Computer Science Northeastern University Shenyang110819 China The College of Information Science and Engineering State Key Laboratory of Synthetical Automation for Process Industries Northeastern University Shenyang110819 China The Faculty of Technology Bielefeld University Bielefeld33619 Germany The Department of Computer Science University of Surrey GuildfordGU2 7XH United Kingdom

In multi-objective evolutionary neural architecture search (NAS), existing predictor-based methods commonly suffer from the rank disorder issue that a candidate high-performance architecture may have a poor ranking compared with the worse architecture in terms of the trained *** alleviate the above issue, we aim to train a Pareto-wise end-to-end ranking classifier to simplify the architecture search process by transforming the complex multi-objective NAS task into a simple classification task. To this end, a classifier-based Pareto evolution approach is proposed, where an online classifier is trained to directly predict the dominance relationship between the candidate and reference architectures. Besides, an adaptive clustering method is designed to select reference architectures for the classifier, and an α-domination assisted approach is developed to address the imbalance issue of positive and negative samples. The proposed approach is compared with a number of state-of-the-art NAS methods on widely-used test datasets, and computation results show that the proposed approach is able to alleviate the rank disorder issue and outperforms other methods. Especially, the proposed method is able to find a set of promising network architectures with different model sizes ranging from 2M to 5M under diverse objectives and constraints. © 2021, CC BY.

关键词： Network architecture

Temperature dependent dielectric spectroscopy of muscle tissue phantom

学校读者我要写书评

暂无评论

Temperature dependent dielectric spectroscopy of muscle tiss...

2019 European Microwave Conference in Central Europe, EuMCE 2019

作者： Fiser, Ondrej Ley, Sebastian Helbig, Marko Sachs, Jurgen Kantova, Michaela Vrba, Jan Department of Biomedical Technology CTU Prague Czech Republic Faculty of Computer Science and Automation TU Ilmenau Germany Department of Electrical Engineering and Information Technology TU Ilmenau Germany Department of Electromagnetic Field CTU Prague Czech Republic

ISBN: (纸本)9782874870675

The temperature dependence of the dielectric parameters of tissues and tissue mimicking phantoms is very important for the non-invasive temperature measurement in medical applications using microwaves. We investigated and proposed a procedure of the measurement of this dependence in the temperature range 25 - 50 °C. Our proposed procedure was verified using distilled water. The results were compared to the literature model in the frequency range 150 - 3000 MHz. Using this method, the temperature dependence of dielectric parameters of new tissue mimicking phantom based on agar, PE powder and TX-151 was measured in the temperature range 25 -50 °C. © 2019 EuMCE.

关键词： Temperature measurement

Data and Physics Driven Learning Models for Fast MRI - Fundamentals and Methodologies from CNN, GAN to Attention and Transformers

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Huang, Jiahao Fang, Yingying Nan, Yang Wu, Huanjun Wu, Yinzhe Gao, Zhifan Li, Yang Wang, Zidong Lio, Pietro Rueckert, Daniel Eldar, Yonina C. Yang, Guang The National Heart and Lung Institute Imperial College London Guy Scadding Building Dovehouse St LondonSW3 6LY United Kingdom The Department of Bioengineering Imperial College London South Kensington LondonSW7 2AZ United Kingdom The School of Biomedical Engineering Sun Yat-Sen University Guangdong Guangzhou510006 China The School of Automation Sciences and Electrical Engineering Beihang University Beijing China The Department of Computer Science Brunel University London LondonUB8 3PH United Kingdom The Computer Laboratory University of Cambridge CambridgeCB3 0FD United Kingdom The Department of Computing Imperial College London South Kensington Campus LondonSW7 2AZ United Kingdom Technical University of Munich Munich80333 Germany Faculty of Mathematics and Computer Science Weizmann Institute of Science Rehovot7610001 Israel

Research studies have shown no qualms about using data driven deep learning models for downstream tasks in medical image analysis, e.g., anatomy segmentation and lesion detection, disease diagnosis and prognosis, and treatment planning. However, deep learning models are not the sovereign remedy for medical image analysis when the upstream imaging is not being conducted properly (with artefacts). This has been manifested in MRI studies, where the scanning is typically slow, prone to motion artefacts, with a relatively low signal to noise ratio, and poor spatial and/or temporal resolution. Recent studies have witnessed substantial growth in the development of deep learning techniques for propelling fast MRI. This article aims to (1) introduce the deep learning based data driven techniques for fast MRI including convolutional neural network and generative adversarial network based methods, (2) survey the attention and transformer based models for speeding up MRI reconstruction, and (3) detail the research in coupling physics and data driven models for MRI acceleration. Finally, we will demonstrate through a few clinical applications, explain the importance of data harmonisation and explainable models for such fast MRI techniques in multicentre and multi-scanner studies, and discuss common pitfalls in current research and recommendations for future research directions. © 2022, CC BY-NC-ND.

关键词： Signal to noise ratio

Short Text Clustering Algorithm Based on Frequent Closed Word Sets

学校读者我要写书评

暂无评论

Short Text Clustering Algorithm Based on Frequent Closed Wor...

International Symposium on Computational Intelligence and Design, ISCID

作者： Chunxia Jin Qiuchan Bai Faculty of Computer and Software Engineering Huaiyin Institute of Technology Huaian China Faculty of Automation Huaiyin Institute of Technology Huaian China

ISBN: (数字)9781728146539

ISBN: (纸本)9781728146546

The text mining of micro-blog topic information can effectively obtain the attention degree of internet users for news events. It is of great significance in the field of public opinion monitoring and analysis. At the situation of the algorithm of traditional frequent word set is suitable for long text information clustering, this paper proposes to mine top-K frequent corpus in short text database and then to divide micro-blog topic texts covering the same frequent word sets into the same cluster. Combined with the largest frequent word-sets for similarity calculation, the overlapped document is re-divided to achieve micro-blog short text clustering. The experimental results of micro-blog topic dataset and the comparison with K-means clustering algorithm show that the proposed algorithm can effectively solve the sparseness and high-dimension problem of micro-blog topic short text clustering and greatly improve the micro-blog short text clustering effect.

关键词： Clustering algorithms Text mining Prediction algorithms Market research Blogs Itemsets

Gesture Interpretation Control System Using Convolutional Neural Networks

学校读者我要写书评

暂无评论

Gesture Interpretation Control System Using Convolutional Ne...

International Conference on Biomedical Innovations and Applications (BIA)

作者： Benedikt Baldursson Behnood Rasti Karl Gudmundsson Dorian Cojocaru Kristinn Andersen Saemundur E. Thorsteinsson Faculty of Electrical and Computer Engineering University of Iceland Hafnarfjordur Iceland Faculty of Automation Computers and Electronics University of Craiova Craiova Romania

ISBN: (纸本)9781728147550

This paper proposes a non-invasive control system for electrical wheelchairs utilizing facial gestures of individuals captured by real-time monocular camera. The images are interpreted with a convolutional neural network that achieves up to ~99.5% overall accuracy. The control system uses an embedded system with a graphics processing unit for predicting real-time throughput with fast inference time. This solution offers great versatility, where the user can make a gesture to depict a command of his choice.

关键词： Non-invasive Gestures Real-time Convolutional neural network Machine learning Gestures control systems Artificial Intelligence Machine Learning Real time Embedded systems Neural network commands

Observer-based controller design for the minimally invasive surgery

学校读者我要写书评

暂无评论

Current Directions in Biomedical engineering 2018年第1期4卷 41-44页

作者： Smolinski, Eike Benkmann, Alexander Drewelow, Wolfgang Jeinsch, Torsten Cappius, Hans-Joachim Westerhoff, Peter Institute of Automation Faculty of Computer Science and Electrical Engineering University of Rostock Rostock Germany World of Medicine GmbH Berlin 10587 Germany

Due to its advantages like minimal trauma and a low risk of infections, the minimally invasive surgery (MIS) has become the standard operation technique for many surgical procedures. During this type of intervention, one of the main tasks from a technical point of view is the control of the pressure in the operation area in order to give the surgeon sufficient space to perform his actions. Because no invasive measurements are available, an extended Kalman filter incorporating the peristaltic behavior of the used medical device was developed to estimate the pressure in the operation area. This contribution focuses on the observer design while a separately designed PI controller using loop shaping was employed for validation. The proposed observer-based control structure was tested with a simulator deploying the principle of physical equivalence. © 2018 Eike Smolinski et al.

关键词： Disturbance rejection

Fast generation of high fidelity RGB-D images by deep-learning with adaptive convolution

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Xian, Chuhua Zhang, Dongjiu Dai, Chengkai Wang, Charlie C.L. Guangdong Provincial Key Lab of Computational Intelligence and Cyberspace Information School of Computer Science and Engineering South China University of Technology Guangzhou Guangzhou510006 China Faculty of Industrial Design Engineering Delft University of Technology Netherlands Department of Mechanical and Automation Engineering Chinese University of Hong Kong Hong Kong

Using the raw data from consumer-level RGB-D cameras as input, we propose a deep-learning based approach to efficiently generate RGB-D images with completed information in high resolution. To process the input images in low resolution with missing regions, new operators for adaptive convolution are introduced in our deep-learning network that consists of three cascaded modules - the completion module, the refinement module and the super-resolution module. The completion module is based on an architecture of encoder-decoder, where the features of input raw RGB-D will be automatically extracted by the encoding layers of a deep neural-network. The decoding layers are applied to reconstruct the completed depth map, which is followed by a refinement module to sharpen the boundary of different regions. For the super-resolution module, we generate RGB-D images in high resolution by multiple layers for feature extraction and a layer for up-sampling. Benefited from the adaptive convolution operators proposed in this paper, our results outperform the existing deep-learning based approaches for RGB-D image complete and super-resolution. As an end-to-end approach, high fidelity RGB-D images can be generated efficiently at the rate of 22 frames per second. Copyright © 2020, The Authors. All rights reserved.

关键词： Cameras