检索结果-内蒙古大学图书馆

30.2 A 22nm 0.26nW/Synapse Spike-Driven Spiking neural Network processing Unit Using Time-Step-First Dataflow and Sparsity-Adaptive In-Memory Computing

30.2 A 22nm 0.26nW/Synapse Spike-Driven Spiking Neural Netwo...

引用

IEEE International conference on Solid-State Circuits (ISSCC)

作者： Ying Liu Yufei Ma Ninghui Shang Tianhao Zhao Peiyu Chen Meng Wu Jiayoon Ru Tianyu Jia Le Ye Zhixuan Wang Ru Huang Peking University Beijing China Nano Core Chip Electronic Technology Hangzhou China Beijing Information Science and Technology University Beijing China

Recently, brain-inspired spiking neural networks (SNNs) have demonstrated tremendous improvement in energy efficiency (EE) and low power by exploiting highly sparse spikes and event-driven design [1–2]. (top of Fig. 30.2.1) With the same spike-based information carrier, the combination of an SNN and dynamic vision sensor (DVS) [3] offers a promising solution for edge AI applications. Typically, the greater the number of synapses in an SNN can improve inference accuracy, but induce more significant power consumption and memory size. Therefore, deploying more synapses in an SNN chip while maintaining low power, memory size and high EE requires co-design of algorithm, architecture and circuits. At the algorithm level, SNN training based on conversion from an artificial NN (ANN) [1] achieves the best accuracy in static image classification, but at the cost of worse EE and higher power consumption, as it requires excessive time steps and energy arising from ANN-SNN conversion. At the architecture level, the multiple-time-step forward computation in SNNs introduces redundant data movement and storage, including input-output spikes, weights, partial sums and membrane potentials, restricting the network scale that can be accommodated by limited on-chip memory, i.e., network density (ND) [4–6]. At the circuit level, in-memory computing (IMC) is attractive to minimize data movement. However, traditional frame-based IMC circuits are incompatible with the highly sparse and fine-grained spike-driven nature of SNNs. Recent advancements in neuromorphic IMC [7] employ integrated-fire (IF) converters, which are globally triggered by power-hungry clocks even if there is no spike, resulting in high power per synapse, i.e., power density (PD).

关键词：

来源：评论

学校读者我要写书评

暂无评论

Edge-AI on Wearable Devices: Myocardial Infarction Detection with Spectrogram and 1D-CNN

Edge-AI on Wearable Devices: Myocardial Infarction Detection...

引用

Mediterranean Electrotechnical conference (MELECON)

作者： Maria Gragnaniello Fiona Balbi Gabriella Martellotta Alessandro Borghese Vincenzo Romano Marrazzo Luca Maresca Giovanni Breglio Andrea Irace Michele Riccio Department of Electrical Engineering and Information Technology (DIETI) University of Naples Federico II Naples Italy

ISBN: (数字)9798350387025

ISBN: (纸本)9798350387032

The Electrocardiogram (ECG) provides a detailed representation of the heart's electrical activity, emerging as a crucial resource for continuous cardiac health monitoring. Recent advances in artificial Intelligence (AI) techniques have revolutionized ECG signal processing, creating new possibilities for every-day health monitoring. The exploitation of these AI technologies has driven a growing interest in wearable devices where the challenge is to implement these functionalities on limited memory resource hardware. In this scenario, the Edge Computing paradigm, where computation occurs near the data source rather than in a remote data center, emerges as a promising solution. This article proposes an efficient approach for Myocardial Infarction (MI) detection based on Deep Learning (DL) methods using spectrogram and a 1D Convolutional neural Network (1D-CNN). The aim is to strike a balance between computational efficiency and accuracy enabling practical application on wearable devices. In the presented work, a study on the impact of the spectrogram parameters and results on the 1D-CNN was conducted. The final training phase yielded a remarkable accuracy of 95.94%, showcasing the efficacy of the proposed approach. Notably, the trained model was successfully deployed on a 32-bit microcontroller featuring an ARM Cortex-M4 architecture, underscoring the feasibility of real-world implementation for embedded systems in healthcare applications.

关键词： Training Accuracy image edge detection Electrocardiography Myocardium Real-time systems Wearable devices

来源：评论

学校读者我要写书评

暂无评论

Voxel-based treatment prediction of glioblastoma multiform tumor using diffusion tensor imaging

Voxel-based treatment prediction of glioblastoma multiform t...

引用

IEEE EMBS International conference on Information Technology applications in Biomedicine (ITAB)

作者： Hadi Sabahi Hamid Soltanian-Zadeh Lisa Scarpace Tom Mikkelsen Control and Intelligent Processing Center of Excellence School of Electrical and Computer Engineering University of Tehran Tehran Iran Image Analysis Laboratory Radiology Department Henry Ford Hospital Detroit MI USA Hermelin Brain Tumor Research Center Neurosurgery Department Henry Ford Health System Detroit MI USA

This paper proposes a method to predict the effect of Bevacizumab therapy on Glioblastoma Multiform (GBM) tumors. The prediction is critical for effective treatment planning. The proposed method is developed and evaluated using Diffusion Tensor Imaging (DTI) and post-contrast T1-weighted Magnetic Resonance images (pc-T1-MRI) of 14 patients with GBM tumors gathered before and after the treatment. First, the proposed method calculates diffusion anisotropy indices (DAI) of all voxels in the brain. These diffusion anisotropy indices are Fractional Anisotropy (FA), Mean Diffusivity (MD), Relative Anisotropy (RA), and Volume Ratio (VR). Then, it registers post-treatment pc-T1-MRI and pre-treatment DAI maps to pre-treatment pc-T1-MRI. Next, it uses a thresholding method to segment the tumor from pc-T1-MRI studies. Comparing Gd-enhanced voxels of the pre- and post-treatment pc-T1-MRI, the DAIs of the tumor are labeled based on their response to the treatment. The voxels of 7 patients are randomly selected to train 4 classifiers (ANN, SVM, KNN, and ANFIS) and then all voxels of the other 7 patients are used to test them. For each classifier, four performance measures (sensitivity, specificity, positive predictive value, and accuracy) are calculated. Experimental results show that the ANFIS is more accurate than the other classifiers in predicting the treatment response.

关键词： Support vector machines Noise Magnetic resonance imaging Sensitivity Tumors Biomedical imaging artificial neural networks

来源：评论

学校读者我要写书评

暂无评论

LVM-Med: learning large-scale self-supervised vision models for medical imaging via second-order graph matching 23

LVM-Med: learning large-scale self-supervised vision models ...

引用

Proceedings of the 37th International conference on neural Information processing Systems

作者： Duy M. H. Nguyen Hoang Nguyen Nghiem T. Diep Tan N. Pham Tri Cao Binh T. Nguyen Paul Swoboda Nhat Ho Shadi Albarqouni Pengtao Xie Daniel Sonntag Mathias Niepert University of Stuttgart and IMPRS for Intelligent Systems and German Research Center for Artificial Intelligence German Research Center for Artificial Intelligence German Research Center for Artificial Intelligence and University of Science - VNUHCM University of Science - VNUHCM Max Planck Institute for Informatics University of Texas at Austin Helmholtz Munich and University Hospital Bonn UC San Diego and MBZUAI German Research Center for Artificial Intelligence and Oldenburg University University of Stuttgart and IMPRS for Intelligent Systems

Obtaining large pre-trained models that can be fine-tuned to new tasks with limited annotated samples has remained an open challenge for medical imaging data. While pre-trained deep networks on imageNet and vision-language foundation models trained on web-scale data are prevailing approaches, their effectiveness on medical tasks is limited due to the significant domain shift between natural and medical images. To bridge this gap, we introduce LVM-Med, the first family of deep networks trained on large-scale medical datasets. We have collected approximately 1.3 million in medical images from 55 publicly available datasets, covering a large number of organs and modalities such as CT, MRI, X-ray, and Ultrasound. We benchmark several state-of-the-art self-supervised algorithms on this dataset and propose a novel self-supervised contrastive learning algorithm using a graph matching formulation. The proposed approach makes three contributions: (i) it integrates prior pair-wise image similarity metrics based on local and global information; (ii) it captures the structural constraints of feature embeddings through a loss function constructed via a combinatorial graph-matching objective; and (iii) it can be trained efficiently end-to-end using modern gradient-estimation techniques for black-box solvers. We thoroughly evaluate the proposed LVM-Med on 15 downstream medical tasks ranging from segmentation and classification to object detection, and both for the in and out-of-distribution settings. LVM-Med empirically outperforms a number of state-of-the-art supervised, self-supervised, and foundation models. For challenging tasks such as Brain Tumor Classification or Diabetic Retinopathy Grading, LVM-Med improves previous vision-language models trained on 1 billion masks by 6-7% while using only a ResNet-50. We release pre-trained models at this link https://***/duyhominhnguyen/LVM-Med.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Paddy Plant Disease Detection using Hybrid Deep Learning Algorithms

Paddy Plant Disease Detection using Hybrid Deep Learning Alg...

引用

International conference of Electronics, Communication and Aerospace Technology (ICECA)

作者： R Suguna S Deepa T Kanimozhi M Chandru G Arun Harshini M Department of Computer Science and Engineering Bannari Amman Institute of Technology Sathyamangalam Erode Tamilnadu India Department of Computer Technology - PG Kongu Engineering College Perundurai Erode Tamilnadu Department of Artificial Intelligence and Data Science Sri Eshwar College of Engineering Coimbatore Department of Computer Science VET-Institute of Arts and Science (Co-Education) College Erode Department of Computer Science and Engineering School of Computing Vel Tech Rangarajan Dr Sagunthala R&D Institute of Science and Technology Chennai M.Sc (Software Systems) Kongu Engineering College Perundurai Erode Tamilnadu

ISBN: (数字)9798350367904

ISBN: (纸本)9798350367911

Agriculture is the backbone of India’s economy, supporting a large portion of the population and establishing the country as a key global player in food production. India ranks as the second-largest producer of both paddy and wheat. However, agricultural productivity is highly susceptible to various plant diseases, which can significantly reduce crop yields and quality. Paddy, one of the most important staple crops, is particularly vulnerable to diseases caused mainly by viral and bacterial related diseases. The climate conditions are also play a vital role in paddy plant diseases. This research work proposes a deep learning-based approach for early detection and classification of paddy diseases. The proposed research devises a hybrid model that combines the VGG19 and ResNet50 algorithms, the system analyses images of both healthy and disease-affected paddy leaves, sourced from the Kaggle repository. This research focuses on predicting four major paddy diseases that can cause yield losses of up to 80%, and in severe cases, can destroy an entire crop: (i) Paddy Blast, (ii) Bacterial Leaf Blight, (iii) Sheath Blight, and (iv) Ufra Disease. The proposed model achieves a high accuracy rate of 98% in predicting paddy diseases and offers a rapid classification process, showcasing the potential of advanced image processing and deep learning techniques to mitigate the impact of paddy plant diseases on agricultural productivity.

关键词： Deep learning Productivity Plant diseases Microorganisms Accuracy Prediction algorithms Real-time systems Classification algorithms System analysis and design Residual neural networks

来源：评论

学校读者我要写书评

暂无评论

dMazeRunner: Executing Perfectly Nested Loops on Dataflow Accelerators

引用

ACM TRANSACTIONS ON EMBEDDED COMPUTING SYSTEMS 2019年第5期18卷 70-70页

作者： Dave, Shail Kim, Youngbin Avancha, Sasikanth Lee, Kyoungwoo Shrivastava, Aviral Arizona State Univ Compiler Microarchitecture Lab Sch Comp Informat & Decis Syst Engn Tempe AZ 85281 USA Yonsei Univ Seoul South Korea Intel Labs Parallel Comp Lab Bangalore Karnataka India

Dataflow accelerators feature simplicity, programmability, and energy-efficiency and are visualized as a promising architecture for accelerating perfectly nested loops that dominate several important applications, including image and media processing and deep learning. Although numerous accelerator designs are being proposed, how to discover the most efficient way to execute the perfectly nested loop of an application onto computational and memory resources of a given dataflow accelerator (execution method) remains an essential and yet unsolved challenge. In this paper, we propose dMazeRunner - to efficiently and accurately explore the vast space of the different ways to spatiotemporally execute a perfectly nested loop on dataflow accelerators (execution methods). The novelty of dMazeRunner framework is in: i) a holistic representation of the loop nests, that can succinctly capture the various execution methods, ii) accurate energy and performance models that explicitly capture the computation and communication patterns, data movement, and data buffering of the different execution methods, and iii) drastic pruning of the vast search space by discarding invalid solutions and the solutions that lead to the same cost. Our experiments on various convolution layers (perfectly nested loops) of popular deep learning applications demonstrate that the solutions discovered by dMazeRunner are on average 9.16x better in Energy-Delay-Product (EDP) and 5.83x better in execution time, as compared to prior approaches. With additional pruning heuristics, dMazeRunner reduces the search time from days to seconds with a mere 2.56% increase in EDP, as compared to the optimal solution.

关键词： Coarse-grained reconfigurable array dataflow deep neural networks loop optimization energy-efficiency systolic arrays mapping analytical model design space exploration

来源：评论

学校读者我要写书评

暂无评论

Classification of Dried Chilli Quality Using image processing

引用

IOP conference Series: Earth and Environmental Science 2021年第1期686卷

作者： Y A Azis N Khuriyati A Suyantohadi Department of Agroindustrial Technology Universitas Gadjah Mada 1st Flora Street Bulaksumur Yogyakarta Indonesia

Chillies are a high-value export commodity. One of the chilli products that has a high export value, such as to the European market, is dried red chillies. Dried chillies' quality is the main parameter that must be maintained if this commodity is to be continuously exported. This study aimed to precisely and accurately classify the quality of dried chillies using physical parameters of the length and color of chillies based on digital image processing. In this research, the quality of dried red chillies was classified using a combination of digital image processing and artificial neural networks (ANN). The quality parameters of dried chillies used as inputs were chilli length, mean energy, mean a*, mean blue, and mean contrast. This study used 150 dried chillies for training data and 36 samples for test data. The classification of the quality of dried red chilli was divided into three classes, which were extra class, class I, and class ii. The result of this study was an artificial neural network structure consisting of 5 input layer cells, 16 hidden layer cells, and two output layer cells. The testing of the system using 36 testing data that determined the values of dried chillies reached 94.4%.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Hyperspectral image classification using neural networks with effect of feature optimization on fused convolutional features

引用

AIP conference Proceedings 2023年第1期2842卷

作者： Rajashree Gadhave R. R. Sedamkar Sujata Alegavi Thakur College of engineering and Technology University of Mumbai Mumbai-400101 Maharashtra India

Analysis of data and synthesis for hyper spectral imaging (HSI) is a new branch of remotely sensed data and planet surveillance technologies. Classification techniques with the help of deep learning for Land cover have recently been a popular research topic, so these techniques are employed in a variety of applications, including agricultural, environmental analysis, military surveillance, mineral extraction and urban exploration. Despite the fact that Convolutional neural networks (CNNs) have attracted a lot of attention and produced impressive results in a range of scene classification tasks, their high computational and storage costs restrict them from being employed in real-time remote sensing applications. Traditional CNN-based techniques, on the other hand, focus on producing scene representation by processing features of original image or from CNN, neglecting the fact that texture pictures or each layer of CNNs convey information that is distinct. To efficiently investigate the characteristics and avoid the above-mentioned drawbacks, a pretrained DNN model-based method is offered. The performance of two different hyperspectral classifiers for land use/land cover classification is compared in this study. The HSI images were classified using Multilayer Perceptron artificial neural networks and Support Vector Machines. The experimental results, which were carried out on the Live Aerial image Hyperspectral Datasets, revealed that classification accuracy is analyzed with and without feature reduction and it is found that with feature reduction performance of proposed algorithm is improved with significant margin with respective to existing aerial image dataset.

关键词：

来源：评论

学校读者我要写书评

暂无评论

image caption generator using CNN & LSTM

引用

AIP conference Proceedings 2023年第1期2754卷

作者： Kodali Anuradha Vellanki Srilakshmi Jampla Sri Naga Sai Siliguri Swapna Konda Keerthi Velmal Lathasri Gangireddi Sadwika Department of CSE Gokaraju Rangaraju Institute of Engineering and Technology Hyderabad India

Computer Vision and Natural Language processing methods have been employed in artificial Intelligence to describe the content of a picture. It is useful in assisting the visually impaired, as well as in self-driving automobiles and other applications. It is not simpleto describe an image, in order to do so, we must use appropriate phrase structure. However, by utilizing deep learning techniques, the process becamesimple and straightforward. CNN (pre-trained Xception model) and LSTM were the algorithms we employed. Convolutional neural networks are deep learning algorithms thatextract features from an input image. The extracted features are sent to the next layer,which is then fed to LSTM. In LSTM, we use a sequence processor to get sentences in order, and then we construct the final caption of the input image through LSTM. It is a type of deep convolutional neural network with depth-wise separable convolutions. We used Flickr_8k *** used Google Colab for training and testing the model. An image will be given as input, the caption will be generated describing the image and the caption is read out(converted to audio) when selected by the user.

关键词：

来源：评论

学校读者我要写书评

暂无评论

A comprehensive review on CNN-based applications for medical imaging classification and segmentation

引用

AIP conference Proceedings 2024年第1期3188卷

作者： Priya Khobragade Ujawalla Gawande Yeshwantrao Chavan College of Engineering (YCCE) Nagpur India

The categorization of medical images, which includes picture segmentation and detection, appears to be a key task in the field of artificial intelligence. Support vector machines (SVM), convolutional neural networks (CNN)have all been applied to the classification of images. Given the capabilities and adaptability of deep learning across a range of machine learning applications, CNN has emerged as one of these methods that has been particularly successful in recent years. This research offers a thorough analysis of CNN’s fundamental ideas and we apply to medical image processing. the core ideas of CNN are outlined, emphasising its capacity to recognise spatial relationships in images and learn hierarchical characteristics. We examine the architecture of CNN and its essential elements, comprising fully connected, pooling, and convolution layers. Further investigation is done into how deep learning can be used to classify medical images more accurately. We offer a variety of CNN models, from basic ones like LeNet-5 to more complex ones like AlexNet, VGGNet, Google Net, and ResNet. Each model’s unique qualities and innovations are discussed, offering insight on how each model contributes to medical picture categorization problems. The application of image processing methods for task augmentation and picture refining is covered in depth in this work. It is addressed how methods like image enhancement, denoising, and feature extraction could enhance the precision and calibre of medical images. Researchers can learn more about the developments and potential of image processing algorithms and CNN-based approaches in the field of medical imaging. This research aims to increase classification accuracy and help the development of reliable and effective medical imaging systems by advancing our understanding of the use of CNN for image processing in relation to applications in medicine.

关键词：

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：