Maintaining filter structure: A Gabor-based convolutional neural network for image analysis

In image segmentation and classification tasks, utilizing filters based on the target object improves performance and requires less training data. We use the Gabor filter as initialization to gain more discriminative power. Considering the mechanism of the error backpropagation procedure to learn the data, after a few updates, filters will lose their initial structure. In this paper, we modify the updating rule in Gradient Descent to maintain the properties of Gabor filters. We use the Left Ventricle (LV) segmentation task and handwritten digit classification task to evaluate our proposed method. We compare Gabor initialization with random initialization and transfer learning initialization using convolutional autoencoders and convolutional networks. We experimented with noisy data and we reduced the amount of training data to compare how different methods of initialization can deal with these matters. The results show that the pixel predictions for the segmentation task are highly correlated with the ground truth. In the classification task, in addition to Gabor and random initialization, we initialized the network using pre-trained weights obtained from a convolutional Autoencoder using two different data sets and pre-trained weights obtained from a convolutional neural network. The experiments confirm the out-performance of Gabor filters comparing to the other initialization method even when using noisy inputs and a lesser amount of training data.     

基于Winograd稀疏算法的卷积神经网络加速器设计与研究

随着卷积神经网络得到愈加广泛的应用,针对其复杂运算的定制硬件加速器得到越来越多的重视与研究。但是,目前定制硬件加速器多采用传统的卷积算法,并且缺乏对神经网络稀疏性的支持,从而丧失了进一步改进硬件,提升硬件性能的空间。重新设计一款卷积神经网络加速器,该加速器基于Winograd稀疏算法,该算法被证明有效降低了卷积神经网络的计算复杂性,并可以很好地适应稀疏神经网络。通过硬件实现该算法,本文的设计可以在减少硬件资源的同时,获得相当大的计算效率。实验表明,相比于传统算法,该加速器设计方案将运算速度提升了近4.15倍;从乘法器利用率的角度出发,相比现有的其他方案,该方案将利用率最多提高了近9倍。     

基于平铺数据流的可配置神经网络加速器

卷积神经网络已经是公认最好的用于深度学习的算法,被广泛地应用于图像识别、自动翻译和广告推荐。由于神经网络结构规模的逐渐增大,使其具有大量的神经元和突触,所以,使用专用加速硬件挖掘神经网络的并行性已经成为了热门的选择。在硬件设计中,经典的平铺结构实现了很高的性能,但是平铺结构的单元利用率很低。目前,随着众多深度学习应用对硬件性能要求的逐渐提高,加速器对单元利用率也具有越来越严格的要求。为了在平铺数据流结构上获得更高的单元利用率,可以调换并行的顺序,采用并行输入特征图和输出通道的方式来提高计算的并行性。但是,随着神经网络运算对硬件性能要求的提高,运算单元阵列必然会越来越大。当阵列大小增加到一定程度,相对单一的并行方式会使利用率逐渐下降。这就需要硬件可以开发更多的神经网络并行度,从而抑制单元空转。同时,为了适应不同的网络结构,要求硬件阵列对神经网络的运算是可配置的。但是,可配置硬件会极大地增加硬件开销和数据的调度难度。提出了一种基于平铺结构加速器的并行度可配置的神经网络加速器。为了减少硬件复杂度,提出了部分配置的技术,既能满足大型单元阵列下单元利用率的提升,也能尽可能地减少硬件额外开销。在阵列大小超过512之后,硬件单元利用率平均可以维持在82%～90%。同时加速器性能与单元阵列数量基本成线性比例上升。     

基于FPGA的卷积神经网络加速器动态余数处理映射模型

将卷积计算转化为矩阵乘法是FPGA上一种高效实现,而现有的转化方法无法根据卷积参数的不同动态调整,限制了卷积计算的并行度.提出一种新的动态余数处理映射模型.该映射模型包含有3个子模型:特征值映射模型,权值映射模型,和输出映射模型.特征值映射模型将特征值转化为特征值矩阵,权值映射模型将权值转化为权值矩阵,特征值矩阵和权值...     

FSS: algorithm and neural network accelerator for style transfer

Neural networks(NNs), owing to their impressive performance, have gradually begun to dominate multimedia processing. For resource-constrained and energy-sensitive mobile devices, an efficient NN accelerator is necessary. Style transfer is an important multimedia application. However, existing arbitrary style transfer networks are complex and not well supported by current NN accelerators, limiting their application on mobile devices. Moreover, the quality of style transfer needs improvement. Thus...     

A genetic programming-based convolutional neural network for image quality evaluations

Neural Computing and Applications - Monitoring the perceptual quality of digital images is fundamentally important since digital image transmissions through the Internet continue to increase...     

A central-point-enhanced convolutional neural network for high-resolution remote-sensing image classification

As one of the most important algorithms in the field of deep learning technology, the convolutional neural network (CNN) has been successfully applied in many fields. CNNs can recognize objects in an image by considering morphology and structure rather than simply individual pixels. One advantage of CNNs is that they exhibit translational invariance; when an image contains a certain degree of distortion or shift, a CNN can still recognize the object in the image. However, this advantage becomes a disadvantage when CNNs are applied to pixel-based classification of remote-sensing images, because their translational invariance characteristics causes distortions in land-cover boundaries and outlines in the classification result image. This problem severely limits the application of CNNs in remote-sensing classification. To solve this problem, we propose a central-point-enhanced convolutional neural network (CE-CNN) to classify high-resolution remote-sensing images. By introducing the central-point-enhanced layer when classifying a sample, the CE-CNN increases the weight of the central point in feather maps while preserving the original textures and characteristics. In our experiment, we selected four representative positions on a high-resolution remote-sensing image to test the classification ability of the proposed method and compared the CE-CNN with the traditional multi-layer perceptron (MLP) and a traditional CNN. The results show that the proposed method can not only achieves a higher classification accuracy but also less distortion and fewer incorrect results at the boundaries of land covers. We further compared the CE-CNN with six state-of-the-art methods: k-NN, maximum likelihood, classification and regression tree (CART), MLP, support vector machine, and CNN. The results show that the CE-CNN’s classification accuracy is better than the other methods.     

基于FPGA的递归神经网络加速器的研究进展

递归神经网络(RNN)近些年来被越来越多地应用在机器学习领域,尤其是在处理序列学习任务中,相比CNN等神经网络性能更为优异。但是RNN及其变体,如LSTM、GRU等全连接网络的计算及存储复杂性较高,导致其推理计算慢,很难被应用在产品中。一方面,传统的计算平台CPU不适合处理RNN的大规模矩阵运算;另一方面,硬件加速平台GPU的共享内存和全局内存使基于GPU的RNN加速器的功耗比较高。FPGA 由于其并行计算及低功耗的特性,近些年来被越来越多地用来做 RNN 加速器的硬件平台。对近些年基于FPGA的RNN加速器进行了研究,将其中用到的数据优化算法及硬件架构设计技术进行了总结介绍,并进一步提出了未来研究的方向。     

Quantum convolutional neural network for image classification

Pattern Analysis and Applications - In this paper we propose two scale-inspired local feature extraction methods based on Quantum Convolutional Neural Network (QCNN) in the Tensorflow quantum...     

A scalable and efficient convolutional neural network accelerator using HLS for a system-on-chip design

This paper presents a configurable convolutional neural network accelerator (CNNA) for a system-on-chip (SoC). The goal was to accelerate inference in different deep learning networks on an embedded SoC platform. The presented CNNA has a scalable architecture that uses high-level synthesis (HLS) and SystemC for the hardware accelerator. It can accelerate any convolutional neural network (CNN) exported from Keras in Python and supports a combination of convolutional, max-pooling, and fully connected layers. A training method with fixed-point quantised weights is proposed and presented in the paper. The CNNA is template-based, enabling it to scale for different targets of the Xilinx Zynq platform. This approach enables design space exploration, which makes it possible to explore several configurations of the CNNA during C and RTL simulation, fitting it to the desired platform and model. The CNN VGG16 was used to test the solution on a Xilinx Ultra96 board using productivity for Zynq (PYNQ). The result gave a high level of accuracy in training with an autoscaled fixed-point Q2.14 format compared to a similar floating-point model. It was able to perform inference in 2.0 s while having an average power consumption of 2.63 W, which corresponds to a power efficiency of 6.0 GOPS/W.     

面向卷积神经网络的高并行度FPGA加速器设计

大多数基于卷积神经网络(CNN)的算法都是计算密集型和存储密集型的,很难应用于具有低功耗要求的航天、移动机器人、智能手机等嵌入式领域.针对这一问题,提出一种面向CNN的高并行度现场可编程逻辑门阵列(FPGA)加速器.首先,比较研究CNN算法中可用于FPGA加速的4类并行度;然后,提出多通道卷积旋转寄存流水(MCRP)结...     

面向云端FPGA的卷积神经网络加速器的设计及其调度

卷积神经网络的高计算复杂性阻碍其广泛用于实时和低功耗应用,现有软件实现方案难以满足其对运算性能与功耗的要求,传统面向FPGA的卷积神经网络构造方式具有流程复杂、周期较长和优化空间较小等问题。针对该问题,根据卷积神经网络计算模式的特点,提出一种面向云端FPGA的卷积神经网络加速器的设计及其调度机制。通过借鉴基于HLS技术、引入循环切割参数和对卷积层循环重排的设计,采用模块化方式构造网络,并进行参数拓展以进一步优化加速器处理过程;通过分析系统任务和资源的特性总结调度方案,且从控制流和数据流两方面对其进行优化设计。与其他已有工作相比,提出的设计提供了一种同时具有灵活性、低能耗、高能效和高性能的解决方案,并且探讨了加速器的高效通用调度方案。实验结果表明,该加速器可在有效提高运算整速度的同时减少功耗。     

A dyadic multi-resolution deep convolutional neural wavelet network for image classification

For almost the past four decades, image classification has gained a lot of attention in the field of pattern recognition due to its application in various fields. Given its importance, several approaches have been proposed up to now. In this paper, we will present a dyadic multi-resolution deep convolutional neural wavelets’ network approach for image classification. This approach consists of performing the classification of one class versus all the other classes of the dataset by the reconstruction of a Deep Convolutional Neural Wavelet Network (DCNWN). This network is based on the Neural Network (NN) architecture, the Fast Wavelet Transform (FWT) and the Adaboost algorithm. It consists, first, of extracting features using the FWT based on the Multi-Resolution Analysis (MRA). These features are used to calculate the inputs of the hidden layer. Second, those inputs are filtered by using the Adaboost algorithm to select the best ones corresponding to each image. Third, we create an AutoEncoder (AE) using wavelet networks of all images. Finally, we apply a pooling for each hidden layer of the wavelet network to obtain a DCNWN that permits the classification of one class and rejects all other classes of the dataset. Classification rates given by our approach show a clear improvement compared to those cited in this article.

     

A brain-like neural network for periodicity analysis.

This paper introduces a brain-like neural model for sound processing. The periodicity analyzing network (PAN) is a bio-inspired neural network of spiking neurons. The PAN consists of complex models of neurons, which can be used for understanding the dynamics of individual neurons and neuronal networks. On a technical level, the PAN is able to compute the ratio of modulation and carrier frequency of harmonic sound signals. The PAN model may, therefore, be used in audio signal processing applications, such as sound source separation, periodicity analysis, and the cocktail party problem.     

一种高效的稀疏卷积神经网络加速器的设计与实现

针对卷积神经网络计算硬件化实现困难的问题,之前大部分卷积神经网络加速器的设计都集中于解决计算性能和带宽瓶颈,忽视了卷积神经网络稀疏性对加速器设计的重要意义,近来少量的能够利用稀疏性的卷积神经网络加速器设计也往往难以同时兼顾计算灵活度、并行效率和资源开销。本文首先比较了不同并行展开方式对利用稀疏性的影响,分析了利用稀疏性的不同方法,然后提出了一种能够利用激活稀疏性加速卷积神经网络计算的同时,相比于同领域其他设计,并行效率更高、额外资源开销更小的并行展开方法,最后完成了这种卷积神经网络加速器的设计并在FPGA上实现。研究结果表明:运行VGG-16网络,在ImageNet数据集下,该并行展开方法实现的稀疏卷积神经网络加速器和使用相同器件的稠密网络设计相比,卷积性能提升了108.8%,整体性能提升了164.6%,具有明显的性能优势。     

Multi-scale convolutional neural network for multi-focus image fusion

In this study, we present new deep learning (DL) method for fusing multi-focus images. Current multi-focus image fusion (MFIF) approaches based on DL methods mainly treat MFIF as a classification task. These methods use a convolutional neural network (CNN) as a classifier to identify pixels as focused or defocused pixels. However, due to unavailability of labeled data to train networks, existing DL-based supervised models for MFIF add Gaussian blur in focused images to produce training data. DL-based unsupervised models are also too simple and only applicable to perform fusion tasks other than MFIF. To address the above issues, we proposed a new MFIF method, which aims to learn feature extraction, fusion and reconstruction components together to produce a complete unsupervised end-to-end trainable deep CNN. To enhance the feature extraction capability of CNN, we introduce a Siamese multi-scale feature extraction module to achieve a promising performance. In our proposed network we applied multiscale convolutions along with skip connections to extract more useful common features from a multi-focus image pair. Instead of using basic loss functions to train the CNN, our model utilizes structure similarity (SSIM) measure as a training loss function. Moreover, the fused images are reconstructed in a multiscale manner to guarantee more accurate restoration of images. Our proposed model can process images with variable size during testing and validation. Experimental results on various test images validate that our proposed method yields better quality fused images that are superior to the fused images generated by compared state-of-the-art image fusion methods.     

Bi-branch deconvolution-based convolutional neural network for image classification

With the rise of deep neural network, convolutional neural networks show superior performances on many different computer vision recognition tasks. The convolution is used as one of the most efficient ways for extracting the details features of an image, while the deconvolution is mostly used for semantic segmentation and significance detection to obtain the contour information of the image and rarely used for image classification. In this paper, we propose a novel network named bi-branch deconvolution-based convolutional neural network (BB-deconvNet), which is constructed by mainly stacking a proposed simple module named Zoom. The Zoom module has two branches to extract multi-scale features from the same feature map. Especially, the deconvolution is borrowed to one of the branches, which can provide distinct features differently from regular convolution through the zoom of learned feature maps. To verify the effectiveness of the proposed network, we conduct several experiments on three object classification benchmarks (CIFAR-10, CIFAR-100, SVHN). The BB-deconvNet shows encouraging performances compared with other state-of-the-art deep CNNs.     

深度卷积神经网络图像语义分割研究进展

在计算机视觉领域中,语义分割是场景解析和行为识别的关键任务,基于深度卷积神经网络的图像语义分割方法已经取得突破性进展。语义分割的任务是对图像中的每一个像素分配所属的类别标签,属于像素级的图像理解。目标检测仅定位目标的边界框,而语义分割需要分割出图像中的目标。本文首先分析和描述了语义分割领域存在的困难和挑战,介绍了语义分割算法性能评价的常用数据集和客观评测指标。然后,归纳和总结了现阶段主流的基于深度卷积神经网络的图像语义分割方法的国内外研究现状,依据网络训练是否需要像素级的标注图像,将现有方法分为基于监督学习的语义分割和基于弱监督学习的语义分割两类,详细阐述并分析这两类方法各自的优势和不足。本文在PASCAL VOC（pattern analysis, statistical modelling and computational learning visual object classes）2012数据集上比较了部分监督学习和弱监督学习的语义分割模型,并给出了监督学习模型和弱监督学习模型中的最优方法,以及对应的MIoU（mean intersection-over-union）。最后,指出了图像语义分割领域未来可能的热点方向。     

关键点图对比图像分类方法

深度学习是目前图像分类的主流方法之一,其重视感受野内的局部信息,却忽略了类别的先验拓扑结构信息。本文提出了一种新的图像分类方法,即Key-D-Graph,这是基于关键点的图对比网络方法,在识别图像类别时可以显式地考虑拓扑先验结构。具体地,图像分类需要2个步骤,第一步是基于关键点构建图像的图表达,即采用深度学习方法识别图像中目标类别的可能关键点,并采用关键点坐标生成图像的拓扑图表达;第二步基于关键点的图像图表达建立图对比网络,以估计待识别图与目标类别之间的结构差异,实现类别判断,该步骤利用了物体的拓扑先验结构信息,实现了基于图像全局结构信息的物体识别。特别的,Key-D-Graph的中间输出结果为类别关键点,具有语义可解释性,便于在实际应用中对算法逐步分析调试。实验结果表明,提出的方法可在效率和精度上超过主流方法,且通过消融实验分析验证了拓扑结构在分类中的作用机制和有效性。