Commenced in January 2007
Frequency: Monthly
Edition: International
Paper Count: 19

Convolutional Neural Networks Related Abstracts

19 2D Convolutional Networks for Automatic Segmentation of Knee Cartilage in 3D MRI

Authors: Ananya Ananya, Karthik Rao


Accurate segmentation of knee cartilage in 3-D magnetic resonance (MR) images for quantitative assessment of volume is crucial for studying and diagnosing osteoarthritis (OA) of the knee, one of the major causes of disability in elderly people. Radiologists generally perform this task in slice-by-slice manner taking 15-20 minutes per 3D image, and lead to high inter and intra observer variability. Hence automatic methods for knee cartilage segmentation are desirable and are an active field of research. This paper presents design and experimental evaluation of 2D convolutional neural networks based fully automated methods for knee cartilage segmentation in 3D MRI. The architectures are validated based on 40 test images and 60 training images from SKI10 dataset. The proposed methods segment 2D slices one by one, which are then combined to give segmentation for whole 3D images. Proposed methods are modified versions of U-net and dilated convolutions, consisting of a single step that segments the given image to 5 labels: background, femoral cartilage, tibia cartilage, femoral bone and tibia bone; cartilages being the primary components of interest. U-net consists of a contracting path and an expanding path, to capture context and localization respectively. Dilated convolutions lead to an exponential expansion of receptive field with only a linear increase in a number of parameters. A combination of modified U-net and dilated convolutions has also been explored. These architectures segment one 3D image in 8 – 10 seconds giving average volumetric Dice Score Coefficients (DSC) of 0.950 - 0.962 for femoral cartilage and 0.951 - 0.966 for tibia cartilage, reference being the manual segmentation.

Keywords: Segmentation, MRI, Convolutional Neural Networks, dilated convolutions, fully automated, knee cartilage, U-net

Procedia PDF Downloads 133
18 Pose Normalization Network for Object Classification

Authors: Bingquan Shen


Convolutional Neural Networks (CNN) have demonstrated their effectiveness in synthesizing 3D views of object instances at various viewpoints. Given the problem where one have limited viewpoints of a particular object for classification, we present a pose normalization architecture to transform the object to existing viewpoints in the training dataset before classification to yield better classification performance. We have demonstrated that this Pose Normalization Network (PNN) can capture the style of the target object and is able to re-render it to a desired viewpoint. Moreover, we have shown that the PNN improves the classification result for the 3D chairs dataset and ShapeNet airplanes dataset when given only images at limited viewpoint, as compared to a CNN baseline.

Keywords: Convolutional Neural Networks, object classification, pose normalization, viewpoint invariant

Procedia PDF Downloads 145
17 Taxonomic Classification for Living Organisms Using Convolutional Neural Networks

Authors: Saed Khawaldeh, Mohamed Elsharnouby, Alaa Eddin Alchalabi, Usama Pervaiz, Tajwar Aleef, Vu Hoang Minh


Taxonomic classification has a wide-range of applications such as finding out more about the evolutionary history of organisms that can be done by making a comparison between species living now and species that lived in the past. This comparison can be made using different kinds of extracted species’ data which include DNA sequences. Compared to the estimated number of the organisms that nature harbours, humanity does not have a thorough comprehension of which specific species they all belong to, in spite of the significant development of science and scientific knowledge over many years. One of the methods that can be applied to extract information out of the study of organisms in this regard is to use the DNA sequence of a living organism as a marker, thus making it available to classify it into a taxonomy. The classification of living organisms can be done in many machine learning techniques including Neural Networks (NNs). In this study, DNA sequences classification is performed using Convolutional Neural Networks (CNNs) which is a special type of NNs.

Keywords: Convolutional Neural Networks, deep networks, taxonomic classification, DNA sequences classification

Procedia PDF Downloads 276
16 A Neural Network Classifier for Identifying Duplicate Image Entries in Real-Estate Databases

Authors: Sergey Ermolin, Olga Ermolin


A Deep Convolution Neural Network with Triplet Loss is used to identify duplicate images in real-estate advertisements in the presence of image artifacts such as watermarking, cropping, hue/brightness adjustment, and others. The effects of batch normalization, spatial dropout, and various convergence methodologies on the resulting detection accuracy are discussed. For comparative Return-on-Investment study (per industry request), end-2-end performance is benchmarked on both Nvidia Titan GPUs and Intel’s Xeon CPUs. A new real-estate dataset from San Francisco Bay Area is used for this work. Sufficient duplicate detection accuracy is achieved to supplement other database-grounded methods of duplicate removal. The implemented method is used in a Proof-of-Concept project in the real-estate industry.

Keywords: Convolutional Neural Networks, visual recognition, triplet loss, spatial batch normalization with dropout, duplicate removal, advertisement technologies, performance benchmarking

Procedia PDF Downloads 193
15 Training a Neural Network to Segment, Detect and Recognize Numbers

Authors: Abhisek Dash


This study had three neural networks, one for number segmentation, one for number detection and one for number recognition all of which are coupled to one another. All networks were trained on the MNIST dataset and were convolutional. It was assumed that the images had lighter background and darker foreground. The segmentation network took 28x28 images as input and had sixteen outputs. Segmentation training starts when a dark pixel is encountered. Taking a window(7x7) over that pixel as focus, the eight neighborhood of the focus was checked for further dark pixels. The segmentation network was then trained to move in those directions which had dark pixels. To this end the segmentation network had 16 outputs. They were arranged as “go east”, ”don’t go east ”, “go south east”, “don’t go south east”, “go south”, “don’t go south” and so on w.r.t focus window. The focus window was resized into a 28x28 image and the network was trained to consider those neighborhoods which had dark pixels. The neighborhoods which had dark pixels were pushed into a queue in a particular order. The neighborhoods were then popped one at a time stitched to the existing partial image of the number one at a time and trained on which neighborhoods to consider when the new partial image was presented. The above process was repeated until the image was fully covered by the 7x7 neighborhoods and there were no more uncovered black pixels. During testing the network scans and looks for the first dark pixel. From here on the network predicts which neighborhoods to consider and segments the image. After this step the group of neighborhoods are passed into the detection network. The detection network took 28x28 images as input and had two outputs denoting whether a number was detected or not. Since the ground truth of the bounds of a number was known during training the detection network outputted in favor of number not found until the bounds were not met and vice versa. The recognition network was a standard CNN that also took 28x28 images and had 10 outputs for recognition of numbers from 0 to 9. This network was activated only when the detection network votes in favor of number detected. The above methodology could segment connected and overlapping numbers. Additionally the recognition unit was only invoked when a number was detected which minimized false positives. It also eliminated the need for rules of thumb as segmentation is learned. The strategy can also be extended to other characters as well.

Keywords: Text Segmentation, Convolutional Neural Networks, OCR, text detection

Procedia PDF Downloads 38
14 Deep Learning Approach to Trademark Design Code Identification

Authors: Girish J. Showkatramani, Arthi M. Krishna, Sashi Nareddi, Naresh Nula, Aaron Pepe, Glen Brown, Greg Gabel, Chris Doninger


Trademark examination and approval is a complex process that involves analysis and review of the design components of the marks such as the visual representation as well as the textual data associated with marks such as marks' description. Currently, the process of identifying marks with similar visual representation is done manually in United States Patent and Trademark Office (USPTO) and takes a considerable amount of time. Moreover, the accuracy of these searches depends heavily on the experts determining the trademark design codes used to catalog the visual design codes in the mark. In this study, we explore several methods to automate trademark design code classification. Based on recent successes of convolutional neural networks in image classification, we have used several different convolutional neural networks such as Google’s Inception v3, Inception-ResNet-v2, and Xception net. The study also looks into other techniques to augment the results from CNNs such as using Open Source Computer Vision Library (OpenCV) to pre-process the images. This paper reports the results of the various models trained on year of annotated trademark images.

Keywords: Convolutional Neural Networks, trademark design code, trademark image classification, trademark image search, Inception-ResNet-v2

Procedia PDF Downloads 55
13 An Accurate Brain Tumor Segmentation for High Graded Glioma Using Deep Learning

Authors: Asad Ali Safi, Sajeeha Ansar, Sheikh Ziauddin, Ahmad R. Shahid, Faraz Ahsan


Gliomas are most challenging and aggressive type of tumors which appear in different sizes, locations, and scattered boundaries. CNN is most efficient deep learning approach with outstanding capability of solving image analysis problems. A fully automatic deep learning based 2D-CNN model for brain tumor segmentation is presented in this paper. We used small convolution filters (3 x 3) to make architecture deeper. We increased convolutional layers for efficient learning of complex features from large dataset. We achieved better results by pushing convolutional layers up to 16 layers for HGG model. We achieved reliable and accurate results through fine-tuning among dataset and hyper-parameters. Pre-processing of this model includes generation of brain pipeline, intensity normalization, bias correction and data augmentation. We used the BRATS-2015, and Dice Similarity Coefficient (DSC) is used as performance measure for the evaluation of the proposed method. Our method achieved DSC score of 0.81 for complete, 0.79 for core, 0.80 for enhanced tumor regions. However, these results are comparable with methods already implemented 2D CNN architecture.

Keywords: Deep learning, Convolutional Neural Networks, brain tumor segmentation, HGG

Procedia PDF Downloads 55
12 LGG Architecture for Brain Tumor Segmentation Using Convolutional Neural Network

Authors: Asad Ali Safi, Sajeeha Ansar, Sheikh Ziauddin, Ahmad R. Shahid, Faraz Ahsan


The most aggressive form of brain tumor is called glioma. Glioma is kind of tumor that arises from glial tissue of the brain and occurs quite often. A fully automatic 2D-CNN model for brain tumor segmentation is presented in this paper. We performed pre-processing steps to remove noise and intensity variances using N4ITK and standard intensity correction, respectively. We used Keras open-source library with Theano as backend for fast implementation of CNN model. In addition, we used BRATS 2015 MRI dataset to evaluate our proposed model. Furthermore, we have used SimpleITK open-source library in our proposed model to analyze images. Moreover, we have extracted random 2D patches for proposed 2D-CNN model for efficient brain segmentation. Extracting 2D patched instead of 3D due to less dimensional information present in 2D which helps us in reducing computational time. Dice Similarity Coefficient (DSC) is used as performance measure for the evaluation of the proposed method. Our method achieved DSC score of 0.77 for complete, 0.76 for core, 0.77 for enhanced tumor regions. However, these results are comparable with methods already implemented 2D CNN architecture.

Keywords: Deep learning, Convolutional Neural Networks, LGG, brain tumor segmentation

Procedia PDF Downloads 48
11 Classification of Computer Generated Images from Photographic Images Using Convolutional Neural Networks

Authors: Chaitanya Chawla, Divya Panwar, Gurneesh Singh Anand, M. P. S Bhatia


This paper presents a deep-learning mechanism for classifying computer generated images and photographic images. The proposed method accounts for a convolutional layer capable of automatically learning correlation between neighbouring pixels. In the current form, Convolutional Neural Network (CNN) will learn features based on an image's content instead of the structural features of the image. The layer is particularly designed to subdue an image's content and robustly learn the sensor pattern noise features (usually inherited from image processing in a camera) as well as the statistical properties of images. The paper was assessed on latest natural and computer generated images, and it was concluded that it performs better than the current state of the art methods.

Keywords: Computer Graphics, classification, Deep learning, Convolutional Neural Networks, Image Forensics

Procedia PDF Downloads 120
10 A Case Study of Deep Learning for Disease Detection in Crops

Authors: Felipe A. Guth, Shane Ward, Kevin McDonnell


In the precision agriculture area, one of the main tasks is the automated detection of diseases in crops. Machine Learning algorithms have been studied in recent decades for such tasks in view of their potential for improving economic outcomes that automated disease detection may attain over crop fields. The latest generation of deep learning convolution neural networks has presented significant results in the area of image classification. In this way, this work has tested the implementation of an architecture of deep learning convolution neural network for the detection of diseases in different types of crops. A data augmentation strategy was used to meet the requirements of the algorithm implemented with a deep learning framework. Two test scenarios were deployed. The first scenario implemented a neural network under images extracted from a controlled environment while the second one took images both from the field and the controlled environment. The results evaluated the generalisation capacity of the neural networks in relation to the two types of images presented. Results yielded a general classification accuracy of 59% in scenario 1 and 96% in scenario 2.

Keywords: Precision Agriculture, Deep learning, Disease Detection, Convolutional Neural Networks

Procedia PDF Downloads 114
9 Using Convolutional Neural Networks to Distinguish Different Sign Language Alphanumerics

Authors: Stephen L. Green, Alexander N. Gorban, Ivan Y. Tyukin


Within the past decade, using Convolutional Neural Networks (CNN)’s to create Deep Learning systems capable of translating Sign Language into text has been a breakthrough in breaking the communication barrier for deaf-mute people. Conventional research on this subject has been concerned with training the network to recognize the fingerspelling gestures of a given language and produce their corresponding alphanumerics. One of the problems with the current developing technology is that images are scarce, with little variations in the gestures being presented to the recognition program, often skewed towards single skin tones and hand sizes that makes a percentage of the population’s fingerspelling harder to detect. Along with this, current gesture detection programs are only trained on one finger spelling language despite there being one hundred and forty-two known variants so far. All of this presents a limitation for traditional exploitation for the state of current technologies such as CNN’s, due to their large number of required parameters. This work aims to present a technology that aims to resolve this issue by combining a pretrained legacy AI system for a generic object recognition task with a corrector method to uptrain the legacy network. This is a computationally efficient procedure that does not require large volumes of data even when covering a broad range of sign languages such as American Sign Language, British Sign Language and Chinese Sign Language (Pinyin). Implementing recent results on method concentration, namely the stochastic separation theorem, an AI system is supposed as an operate mapping an input present in the set of images u ∈ U to an output that exists in a set of predicted class labels q ∈ Q of the alphanumeric that q represents and the language it comes from. These inputs and outputs, along with the interval variables z ∈ Z represent the system’s current state which implies a mapping that assigns an element x ∈ ℝⁿ to the triple (u, z, q). As all xi are i.i.d vectors drawn from a product mean distribution, over a period of time the AI generates a large set of measurements xi called S that are grouped into two categories: the correct predictions M and the incorrect predictions Y. Once the network has made its predictions, a corrector can then be applied through centering S and Y by subtracting their means. The data is then regularized by applying the Kaiser rule to the resulting eigenmatrix and then whitened before being split into pairwise, positively correlated clusters. Each of these clusters produces a unique hyperplane and if any element x falls outside the region bounded by these lines then it is reported as an error. As a result of this methodology, a self-correcting recognition process is created that can identify fingerspelling from a variety of sign language and successfully identify the corresponding alphanumeric and what language the gesture originates from which no other neural network has been able to replicate.

Keywords: Deep learning, Sign Language, Convolutional Neural Networks, shallow correctors

Procedia PDF Downloads 14
8 MITOS-RCNN: Mitotic Figure Detection in Breast Cancer Histopathology Images Using Region Based Convolutional Neural Networks

Authors: Siddhant Rao


Studies estimate that there will be 266,120 new cases of invasive breast cancer and 40,920 breast cancer induced deaths in the year of 2018 alone. Despite the pervasiveness of this affliction, the current process to obtain an accurate breast cancer prognosis is tedious and time consuming. It usually requires a trained pathologist to manually examine histopathological images and identify the features that characterize various cancer severity levels. We propose MITOS-RCNN: a region based convolutional neural network (RCNN) geared for small object detection to accurately grade one of the three factors that characterize tumor belligerence described by the Nottingham Grading System: mitotic count. Other computational approaches to mitotic figure counting and detection do not demonstrate ample recall or precision to be clinically viable. Our models outperformed all previous participants in the ICPR 2012 challenge, the AMIDA 2013 challenge and the MITOS-ATYPIA-14 challenge along with recently published works. Our model achieved an F- measure score of 0.955, a 6.11% improvement in accuracy from the most accurate of the previously proposed models.

Keywords: Machine Learning, Breast Cancer, Convolutional Neural Networks, mitotic count

Procedia PDF Downloads 105
7 Automatic Measurement of Garment Sizes Using Deep Learning

Authors: Maulik Parmar, Sumeet Sandhu


The online fashion industry experiences high product return rates. Many returns are because of size/fit mismatches -the size scale on labels can vary across brands, the size parameters may not capture all fit measurements, or the product may have manufacturing defects. Warehouse quality check of garment sizes can be semi-automated to improve speed and accuracy. This paper presents an approach for automatically measuring garment sizes from a single image of the garment -using Deep Learning to learn garment keypoints. The paper focuses on the waist size measurement of jeans and can be easily extended to other garment types and measurements. Experimental results show that this approach can greatly improve the speed and accuracy of today’s manual measurement process.

Keywords: Deep learning, Convolutional Neural Networks, distortion, image warping, garment measurements, keypoints

Procedia PDF Downloads 16
6 Performance Evaluation of Distributed Deep Learning Frameworks in Cloud Environment

Authors: Shuen-Tai Wang, Chau-Yi Chou, Fang-An Kuo, Yu-Bin Fang


2016 has become the year of the Artificial Intelligence explosion. AI technologies are getting more and more matured that most world well-known tech giants are making large investment to increase the capabilities in AI. Machine learning is the science of getting computers to act without being explicitly programmed, and deep learning is a subset of machine learning that uses deep neural network to train a machine to learn  features directly from data. Deep learning realizes many machine learning applications which expand the field of AI. At the present time, deep learning frameworks have been widely deployed on servers for deep learning applications in both academia and industry. In training deep neural networks, there are many standard processes or algorithms, but the performance of different frameworks might be different. In this paper we evaluate the running performance of two state-of-the-art distributed deep learning frameworks that are running training calculation in parallel over multi GPU and multi nodes in our cloud environment. We evaluate the training performance of the frameworks with ResNet-50 convolutional neural network, and we analyze what factors that result in the performance among both distributed frameworks as well. Through the experimental analysis, we identify the overheads which could be further optimized. The main contribution is that the evaluation results provide further optimization directions in both performance tuning and algorithmic design.

Keywords: Artificial Intelligence, Machine Learning, Deep learning, Convolutional Neural Networks

Procedia PDF Downloads 4
5 Comparison of Classical Computer Vision vs. Convolutional Neural Networks Approaches for Weed Mapping in Aerial Images

Authors: Paulo Cesar Pereira Junior, Alexandre Monteiro, Rafael da Luz Ribeiro, Antonio Carlos Sobieranski, Aldo von Wangenheim


In this paper, we present a comparison between convolutional neural networks and classical computer vision approaches, for the specific precision agriculture problem of weed mapping on sugarcane fields aerial images. A systematic literature review was conducted to find which computer vision methods are being used on this specific problem. The most cited methods were implemented, as well as four models of convolutional neural networks. All implemented approaches were tested using the same dataset, and their results were quantitatively and qualitatively analyzed. The obtained results were compared to a human expert made ground truth for validation. The results indicate that the convolutional neural networks present better precision and generalize better than the classical models.

Keywords: Digital Image Processing, Precision Agriculture, Deep learning, Unmanned Aerial Vehicles, Convolutional Neural Networks, semantic segmentation

Procedia PDF Downloads 1
4 Visual Inspection of Road Conditions Using Deep Convolutional Neural Networks

Authors: Christos Theoharatos, Dimitris Tsourounis, Spiros Oikonomou, Andreas Makedonas


This paper focuses on the problem of visually inspecting and recognizing the road conditions in front of moving vehicles, targeting automotive scenarios. The goal of road inspection is to identify whether the road is slippery or not, as well as to detect possible anomalies on the road surface like potholes or body bumps/humps. Our work is based on an artificial intelligence methodology for real-time monitoring of road conditions in autonomous driving scenarios, using state-of-the-art deep convolutional neural network (CNN) techniques. Initially, the road and ego lane are segmented within the field of view of the camera that is integrated into the front part of the vehicle. A novel classification CNN is utilized to identify among plain and slippery road textures (e.g., wet, snow, etc.). Simultaneously, a robust detection CNN identifies severe surface anomalies within the ego lane, such as potholes and speed bumps/humps, within a distance of 5 to 25 meters. The overall methodology is illustrated under the scope of an integrated application (or system), which can be integrated into complete Advanced Driver-Assistance Systems (ADAS) systems that provide a full range of functionalities. The outcome of the proposed techniques present state-of-the-art detection and classification results and real-time performance running on AI accelerator devices like Intel’s Myriad 2/X Vision Processing Unit (VPU).

Keywords: Embedded Systems, Deep learning, Convolutional Neural Networks, road condition classification

Procedia PDF Downloads 1
3 Fall Detection Using Convolutional Neural Network Classification of Movement Data

Authors: Jalal Alizadeh, Christopher Fricke, Martin Bogdan, Joseph Classen


Severe falls are a major cause of morbidity and mortality in neurodegenerative disorders. A promising approach to differentiate between normal daily activity and fall events is based on classification of movement data collected by accelerometers and gyroscopes using deep learning methods. We aimed to design a classifier based on convolutional neural networks to detect fall events in a large dataset of young participants imitating falls. In addition, we validated the classifier in a second dataset including real falls of elderly people. In order to develop the classifier, we used two public domain datasets: SisFall and Farseeing. The SisFall dataset consists of 23 subjects performing 19 different activities of daily living (ADL) and 15 imitated falls, each with five repetitions. The Farseeing dataset consists of real fall events that happened to 22 elderly people. In the SisFall dataset, movement data was recorded with two 3D accelerometers at sample rate of 200 Hz. On the other hand, the Farseeing dataset was collected with two 3D accelerometers worn on lower back and tight at frequency sample of 100 Hz and 20 Hz, respectively, and only the movement data corresponding to 100 Hz were used. Accelerometer data were split into two groups (“ADL” and “FALL”), epoched into 2.5 seconds segments and the root sum squared of the three data channels was computed for further analysis. To make our classifier more reliable, we excluded randomly three subjects from each group for SisFall-validation and then split the remaining dataset into training and testing data with a split ratio of 70/30. Next, we performed 10-fold cross correlation and averaged the results. Afterwards, we calculated the accuracy, precision and F1 score (a combination of precision and sensitivity metrics) of the classifier. Finally, the classifier was validated on the Farseeing dataset. The accuracies of our classifier were ~93% for the test and the SisFall-validation sets using the SisFall data. Validating the trained SisFall network on the Farseeing dataset yielded an accuracy of ~78%. Accordingly, the F1 score results were ~73% for test, ~73% for SisFall-validation and ~74% for Farseeing-validation sets. In addition, the averaged precision was ~92% for the Farseeing-validation set. These results suggest that the proposed classifier can reliably distinguish a fall event from a normal activity. The results were robust and were reproduced for 10-fold cross correlation as proven by the small variance of the accuracy distribution. The high precision suggests that the algorithm has the ability to reduce the incidence of false alarms.

Keywords: Deep learning, Accelerometers, Convolutional Neural Networks, human fall detection

Procedia PDF Downloads 1
2 Normal and Peaberry Coffee Beans Classification from Green Coffee Bean Images Using Convolutional Neural Networks and Support Vector Machine

Authors: Hira Lal Gope, Hidekazu Fukai


The aim of this study is to develop a system which can identify and sort peaberries automatically at low cost for coffee producers in developing countries. In this paper, the focus is on the classification of peaberries and normal coffee beans using image processing and machine learning techniques. The peaberry is not bad and not a normal bean. The peaberry is born in an only single seed, relatively round seed from a coffee cherry instead of the usual flat-sided pair of beans. It has another value and flavor. To make the taste of the coffee better, it is necessary to separate the peaberry and normal bean before green coffee beans roasting. Otherwise, the taste of total beans will be mixed, and it will be bad. In roaster procedure time, all the beans shape, size, and weight must be unique; otherwise, the larger bean will take more time for roasting inside. The peaberry has a different size and different shape even though they have the same weight as normal beans. The peaberry roasts slower than other normal beans. Therefore, neither technique provides a good option to select the peaberries. Defect beans, e.g., sour, broken, black, and fade bean, are easy to check and pick up manually by hand. On the other hand, the peaberry pick up is very difficult even for trained specialists because the shape and color of the peaberry are similar to normal beans. In this study, we use image processing and machine learning techniques to discriminate the normal and peaberry bean as a part of the sorting system. As the first step, we applied Deep Convolutional Neural Networks (CNN) and Support Vector Machine (SVM) as machine learning techniques to discriminate the peaberry and normal bean. As a result, better performance was obtained with CNN than with SVM for the discrimination of the peaberry. The trained artificial neural network with high performance CPU and GPU in this work will be simply installed into the inexpensive and low in calculation Raspberry Pi system. We assume that this system will be used in under developed countries. The study evaluates and compares the feasibility of the methods in terms of accuracy of classification and processing speed.

Keywords: Sorting, Convolutional Neural Networks, support vector machine, coffee bean, peaberry

Procedia PDF Downloads 1
1 A Survey of Field Programmable Gate Array Based Convolutional Neural Networks Accelerators

Authors: Wei Zhang


With the rapid development of deep learning, neural network and deep learning algorithms play a significant role in various practical applications. Due to the high accuracy and good performance, especially CNNs, it has become a research hot spot in research organizations for the past few years. However, the size of the networks becomes increasingly large scale due to the demands of the practical applications, which poses a significant challenge to construct a high-performance implementation of deep learning neural networks. Meanwhile, many of these application scenarios also have strict requirements on the performance and low-power consumption of hardware devices. Therefore, it is particularly critical to choose a moderate computing platform for hardware acceleration of CNNs. This article is aimed to survey the recent advance in FPGA-based acceleration of CNNs. Various designs and implementations of the accelerator based on FPGA under different devices and network models are overviewed, and the versions of GPUs, ASICs (Application-Specific Integrated Circuits), and DSPs (Digital Signal Processors) are compared, which is to present our own critical analysis and comments. Finally, we give a discussion on different perspectives of these acceleration and optimization methods on FPGA platforms and to further explore the opportunities and challenges for future research. More helpfully, we give a prospect for future development of FPGA-based accelerator.

Keywords: FPGA, Deep learning, Convolutional Neural Networks, field programmable gate array, CNN, hardware accelerator

Procedia PDF Downloads 1