基于深度学习的铁路图像场景分类优化研究

铁路检测、监测领域产生海量的图像数据,基于图像场景进行分类对图像后续分析、管理具有重要价值.本文提出一种结合深度卷积神经神经网络DCNN （Deep Convolutional Neural Networks）与梯度类激活映射Grad-CAM （Grad Class Activation Mapping）的可视化场景分类模型,DCNN在铁路场景分类图像数据集进行迁移学习,实现特征提取,Grad-CAM根据梯度全局平均计算权重实现对类别的加权热力图及激活分数计算,提升分类模型可解释性.实验中对比了不同的DCNN网络结构对铁路图像场景分类任务性能影响,对场景分类模型实现可视化解释,基于可视化模型提出了通过降低数据集内部偏差提升模型分类能力的优化流程,验证了深度学习技术对于图像场景分类任务的有效性.     

基于卷积神经网络的直线描述方法研究

为提高直线特征匹配的可靠性,提出一种基于卷积神经网络（CNN）学习的直线特征描述方法。构建用于网络学习的大规模直线数据集,该数据集包含约20.8万对匹配直线对,每条直线用其周围的局部图像块表征。将图像块输入CNN,利用HardNet网络结构提取特征,使用三元组损失函数进行训练,输出强鲁棒性的直线特征描述子。实验结果表明,与手工设计的描述子MSLD和IOCD相比,该描述子在视角、模糊、尺度和旋转变化下均具有较好的区分性,在图像拼接应用中同样表现出良好的描述性能。     

Caffe CNN-based classification of hyperspectral images on GPU

Deep learning techniques based on Convolutional Neural Networks (CNNs) are extensively used for the classification of hyperspectral images. These techniques present high computational cost. In this paper, a GPU (Graphics Processing Unit) implementation of a spatial-spectral supervised classification scheme based on CNNs and applied to remote sensing datasets is presented. In particular, two deep learning libraries, Caffe and CuDNN, are used and compared. In order to achieve an efficient GPU projection, different techniques and optimizations have been applied. The implemented scheme comprises Principal Component Analysis (PCA) to extract the main features, a patch extraction around each pixel to take the spatial information into account, one convolutional layer for processing the spectral information, and fully connected layers to perform the classification. To improve the initial GPU implementation accuracy, a second convolutional layer has been added. High speedups are obtained together with competitive classification accuracies.

     

Applying Wide & Deep Learning Model for Android Malware Classification

Android malware has exploded in popularity in recent years, due to the platform’s dominance of the mobile market. With the advancement of deep learning technology, numerous deep learning-based works have been proposed for the classification of Android malware. Deep learning technology is designed to handle a large amount of raw and continuous data, such as image content data. However, it is incompatible with discrete features, i.e., features gathered from multiple sources. Furthermore, if the feature set is already well-extracted and sparsely distributed, this technology is less effective than traditional machine learning. On the other hand, a wide learning model can expand the feature set to enhance the classification accuracy. To maximize the benefits of both methods, this study proposes combining the components of deep learning based on multi-branch CNNs (Convolutional Network Neural) with wide learning method. The feature set is evaluated and dynamically partitioned according to its meaning and generalizability to subsets when used as input to the model’s wide or deep component. The proposed model, partition, and feature set quality are all evaluated using the K-fold cross validation method on a composite dataset with three types of features: API, permission, and raw image. The accuracy with Wide and Deep CNN (WDCNN) model is 98.64%, improved by 1.38% compared to RNN (Recurrent Neural Network) model.     

改进LeNet-5网络在图像分类中的应用

LeNet-5卷积神经网络（CNN）虽然在手写数字识别上取得很好的分类效果,但在具有复杂纹理特征的数据集上分类精度不高。为提高网络在复杂纹理特征图像上分类的正确率,提出一种改进的LeNet-5网络结构。引入跨连思想,充分利用网络提取的低层次特征;把Inception V1模块嵌入LeNet-5卷积神经网络,提取图像的多尺度特征;输出层使用softmax函数对图像进行分类。在Cifar-10和Fashion MNIST数据集上进行的实验结果表明,改进的卷积神经网络在复杂纹理特征数据集上具有很好的分类能力。     

Combining additive input noise annealing and pattern transformations for improved handwritten character recognition

Two problems that burden the learning process of Artificial Neural Networks with Back Propagation are the need of building a full and representative learning data set, and the avoidance of stalling in local minima. Both problems seem to be closely related when working with the handwritten digits contained in the MNIST dataset. Using a modest sized ANN, the proposed combination of input data transformations enables the achievement of a test error as low as 0.43%, which is up to standard compared to other more complex neural architectures like Convolutional or Deep Neural Networks.     

基于注意力机制网络的航运监控图像识别模型

针对已有的航运监控图像识别模型C3D里中级表征学习能力有限,有效特征的提取容易受到噪声的干扰,且特征的提取忽视了整体特征与局部特征之间关系的问题,提出了一种新的基于注意力机制网络的航运监控图像识别模型。该模型基于卷积神经网络（CNN）框架,首先,通过特征提取器提取图像的浅层次特征;然后,基于CNN对不同区域激活特征的不同响应强度,生成注意力信息并实现对局部判别性特征的提取;最后,使用多分支的CNN结构融合局部判别性特征和图像全局纹理特征,从而利用局部判别性特征和图像全局纹理特征的交互关系提升CNN学习中级表征的能力。实验结果表明,所提出的模型在航运图像数据集上的识别准确率达到91.8%,相较于目前的C3D模型提高了7.2个百分点,相较于判别滤波器组卷积神经网络（DFL-CNN）模型提高了0.6个百分点。可见所提模型能够准确判断船舶的状态,可以有效应用于航运监控项目。     

Deep Convolutional Neural Network (Falcon) and transfer learning‐based approach to detect malarial parasite

Deep learning models have already benchmarked its demonstration in the applications of Medical Sciences. Present day medical industries suffer due to deadly disease such as malaria etc. As per the report from World Health Organization (WHO), it is noted that the amount of caution and care taken per patient by a human doctor to cure malaria is decreasing. To address this issue, this paper proposes an automated solution for the detection of malaria from the real-time image. The key idea of the proposed solution is to use a Deep Convolutional Neural Network (DCNN) called “Falcon” to detect the parasitic cells from blood smeared slide images of Malaria Screener. Furthermore, the class accuracy of the given dataset samples is maintained in order to model not only the normal case but to accurately predict the presence of malaria as well. Experimental results confirms that the model does not possess overfitting, class imbalance, and provides a reasonable classification report and trustworthy accuracy with 95.2?% when compared to the state-of-the-art Convolutional Neural Network (CNN) models.

     

Learning Sparse Feature Representations Using Probabilistic Quadtrees and Deep Belief Nets

Learning sparse feature representations is a useful instrument for solving an unsupervised learning problem. In this paper, we present three labeled handwritten digit datasets, collectively called n-MNIST by adding noise to the MNIST dataset, and three labeled datasets formed by adding noise to the offline Bangla numeral database. Then we propose a novel framework for the classification of handwritten digits that learns sparse representations using probabilistic quadtrees and Deep Belief Nets. On the MNIST, n-MNIST and noisy Bangla datasets, our framework shows promising results and outperforms traditional Deep Belief Networks.     

基于深度卷积神经网络的图像检索算法研究

为解决卷积神经网络在提取图像特征时所造成的特征信息损失,提高图像检索的准确率,提出了一种基于改进卷积神经网络LeNet-L的图像检索算法。首先,改进LeNet-5卷积神经网络结构,增加网络结构深度。然后,对深度卷积神经网络模型LeNet-L进行预训练,得到训练好的网络模型,进而提取出图像高层语义特征。最后,通过距离函数比较待检图像与图像库的相似度,得出相似图像。在Corel数据集上,与原模型以及传统的SVM主动学习图像检索方法相比,该图像检索方法有较高的准确性。经实验结果表明,改进后的卷积神经网络具有更好的检索效果。     

深度先验图像特征在城市遥感大数据中的应用

图像特征提取始终是计算机视觉和图像处理的核心任务.随着深度学习的快速发展,卷积神经网络逐渐取代传统图像特征算子,成为特征提取的主要算法.本文针对城市遥感数据众包标记系统中的数据关联问题,结合卷积神经网络和池化编码,提出基于深度先验的图像特征提取方法.该特征能有效聚焦室外图像近处物体,并通过图像检索实验验证了其对室外图像的良好表征能力.     

A novel group squeeze excitation sparsely connected convolutional networks for SAR target classification

Automatic Target Recognition (ATR) based on Synthetic Aperture Radar (SAR) images plays a key role in military applications. However, there are difficulties with this traditional recognition method. Principally, it is a challenge to design robust features and classifiers for different SAR images. Although Convolutional Neural Networks (CNNs) are very successful in many image classification tasks, building a deep network with limited labeled data remains a problem. The topologies of CNNs like the fully connected structure will lead to redundant parameters and the negligence of channel-wise information flow. A novel CNNs approach, called Group Squeeze Excitation Sparsely Connected Convolutional Networks (GSESCNNs), is therefore proposed as a solution. The group squeeze excitation performs dynamic channel-wise feature recalibration with less parameters than squeeze excitation. Sparsely connected convolutional networks are a more efficient way to operate the concatenation of feature maps from different layers. Experimental results on Moving and Stationary Target Acquisition and Recognition (MSTAR) SAR images, demonstrate that this approach achieves, at 99.79%, the best prediction accuracy, outperforming the most common skip connection models, such as Residual Networks and Densely Connected Convolutional Networks, as well as other methods reported in the MSTAR dataset.     

Deep feature learning versus shallow feature learning systems for joint use of airborne thermal hyperspectral and visible remote sensing data

Recently, the development of various remote sensing sensors has provided more reliable information and data for identification of different ground classes. Accordingly, multisensory fusion techniques are applied to enhance the process of information extraction from complementary airborne and spaceborne remote sensing data. Most of previous research in the literature has focused on the extraction of shallow features from a specific sensor and on classification of the resulted feature space using decision fusion systems. In recent years, Deep Learning (DL) algorithms have drawn a lot of attention in the machine learning area and have had different remote sensing applications, especially on data fusion. This study presents two different feature-learning strategies for the fusion of hyperspectral thermal infrared (HTIR) and visible remote sensing data. First, a Deep Convolutional Neural Network (DCNN)-Support Vector Machine (SVM) was utilized on the features of two datasets to provide the class labels. To validate the results with other learning strategies, a shallow feature model was used, as well. This model was based on feature fusion and decision fusion that classified and fused the two datasets. A co-registered thermal infrared hyperspectral (HTIR) and Fine Resolution Visible (Vis) RGB imagery was available from Quebec of Canada to examine the effectiveness of the proposed method. Experimental results showed that, except for the computational time, the proposed deep learning model outperformed shallow feature-based strategies in the classification performance that was based on its accuracy.     

基于深度学习的SIFT图像检索算法

深度学习作为一个新的机器学习方向,被应用到计算机视觉领域上成效显著.为了解决分布式的尺度不变特征转换（Scale-Invariant Feature Transform,SIFT）算法效率低和图像特征提取粗糙问题,提出一种基于深度学习的SIFT图像检索算法.算法思想：在Spark平台上,利用深度卷积神经网络（Convolutional Neural Network,CNN）模型进行SIFT特征抽取,再利用支持向量机（Support Vector Machine,SVM）对图像库进行无监督聚类,然后再利用自适应的图像特征度量来对检索结果进行重排序,以改善用户体验.在Corel图像集上的实验结果显示,与传统SIFT算法相比,基于深度学习的SIFT图像检索算法的查准率和查全率大约提升了30个百分点,检索效率得到了提高,检索结果图像排序也得到了优化.     

数据异常情况下遥感影像时间序列分类算法

针对时序遥感图像数据异常时卷积神经网络对其分类性能较差的问题,提出了一种端到端的多模式与多单模架构相结合的网络结构。首先,通过多元时序模型和单变量时间序列模型对多维时间序列进行多尺度特征提取;然后,基于像素空间坐标信息,通过自动编码形式完成遥感图像的时空序列特征的构建;最后,通过全连接层和softmax函数实现分类。在数据异常（数据缺失和数据扭曲）的情况下,提出的算法和一维卷积神经网络（1D-CNN）、多通道深度神经网络（MCDNN）、时序卷积神经网络（TSCNN）和长短期记忆（LSTM）网络等通用时间序列遥感影像分类算法进行分析比较。实验结果表明,所提的利用端到端的多模式与多单模式架构融合的网络在数据异常的情况下分类精度最高,F1值达到了93.40%。     

Automatic feature extraction of time-series applied to fault severity assessment of helical gearbox in stationary and non-stationary speed operation

Signals captured in rotating machines to obtain the status of their components can be considered as a source of massive information. In current methods based on artificial intelligence to fault severity assessment, features are first generated by advanced signal processing techniques. Then feature selection takes place, often requiring human expertise. This approach, besides time-consuming, is highly dependent on the machinery configuration as in general the results obtained for a mechanical system cannot be reused by other systems. Moreover, the information about time events is often lost along the process, preventing the discovery of faulty state patterns in machines operating under time-varying conditions. In this paper a novel method for automatic feature extraction and estimation of fault severity is proposed to overcome the drawbacks of classical techniques. The proposed method employs a Deep Convolutional Neural Network pre-trained by a Stacked Convolutional Autoencoder. The robustness and accuracy of this new method are validated using a dataset with different severity conditions on failure mode in a helical gearbox, working in both constant and variable speed of operation. The results show that the proposed unsupervised feature extraction method is effective for the estimation of fault severity in helical gearbox, and it has a consistently better performance in comparison with other reported feature extraction methods.     

Implementation of Hybrid Deep Reinforcement Learning Technique for Speech Signal Classification

Classification of speech signals is a vital part of speech signal processing systems. With the advent of speech coding and synthesis, the classification of the speech signal is made accurate and faster. Conventional methods are considered inaccurate due to the uncertainty and diversity of speech signals in the case of real speech signal classification. In this paper, we use efficient speech signal classification using a series of neural network classifiers with reinforcement learning operations. Prior classification of speech signals, the study extracts the essential features from the speech signal using Cepstral Analysis. The features are extracted by converting the speech waveform to a parametric representation to obtain a relatively minimized data rate. Hence to improve the precision of classification, Generative Adversarial Networks are used and it tends to classify the speech signal after the extraction of features from the speech signal using the cepstral coefficient. The classifiers are trained with these features initially and the best classifier is chosen to perform the task of classification on new datasets. The validation of testing sets is evaluated using RL that provides feedback to Classifiers. Finally, at the user interface, the signals are played by decoding the signal after being retrieved from the classifier back based on the input query. The results are evaluated in the form of accuracy, recall, precision, f-measure, and error rate, where generative adversarial network attains an increased accuracy rate than other methods: Multi-Layer Perceptron, Recurrent Neural Networks, Deep belief Networks, and Convolutional Neural Networks.     

Deep residual learning for image steganalysis

Image steganalysis is to discriminate innocent images and those suspected images with hidden messages. This task is very challenging for modern adaptive steganography, since modifications due to message hiding are extremely small. Recent studies show that Convolutional Neural Networks (CNN) have demonstrated superior performances than traditional steganalytic methods. Following this idea, we propose a novel CNN model for image steganalysis based on residual learning. The proposed Deep Residual learning based Network (DRN) shows two attractive properties than existing CNN based methods. First, the model usually contains a large number of network layers, which proves to be effective to capture the complex statistics of digital images. Second, the residual learning in DRN preserves the stego signal coming from secret messages, which is extremely beneficial for the discrimination of cover images and stego images. Comprehensive experiments on standard dataset show that the DRN model can detect the state of arts steganographic algorithms at a high accuracy. It also outperforms the classical rich model method and several recently proposed CNN based methods.     

A deep neural network based classifier for brain tumor diagnosis

Classification process plays a key role in diagnosing brain tumors. Earlier research works are intended for identifying brain tumors using different classification techniques. However, the False Alarm Rates (FARs) of existing classification techniques are high. To improve the early-stage brain tumor diagnosis via classification the Weighted Correlation Feature Selection Based Iterative Bayesian Multivariate Deep Neural Learning (WCFS-IBMDNL) technique is proposed in this work. The WCFS-IBMDNL algorithm considers medical dataset for classifying the brain tumor diagnosis at an early stage. At first, the WCFS-IBMDNL technique performs Weighted Correlation-Based Feature Selection (WC-FS) by selecting subsets of medical features that are relevant for classification of brain tumors. After completing the feature selection process, the WCFS-IBMDNL technique uses Iterative Bayesian Multivariate Deep Neural Network (IBMDNN) classifier for reducing the misclassification error rate of brain tumor identification. The WCFS-IBMDNL technique was evaluated in JAVA language using Disease Diagnosis Rate (DDR), Disease Diagnosis Time (DDT), and FAR parameter through the epileptic seizure recognition dataset.     

Deep Convolutional Neural Networks with Spatial Regularization,Volume and Star-Shape Priors for Image Segmentation

Journal of Mathematical Imaging and Vision - Deep Convolutional Neural Networks (DCNNs) can well extract the features from natural images. However, the classification functions in the existing...