基于卷积神经网络特征融合的人脸识别算法

为提高卷积神经网络的识别性能,提出了一种基于多种卷积神经网络模型的特征融合方法。论文通过构建一个深度学习网络,将多种卷积神经网络模型如ResNet、InceptionV3和VGG19提取的特征进行融合,并将融合后的特征应用到人脸识别中,据此训练出特征融合网络模型的网络参数;最后利用计算求出的阈值来区分类别。实验结果表明,在人脸库LFW数据集上,论文算法的人脸识别率可达98%;与现有的单一卷积神经网络相比,论文算法识别率更高。     

基于对抗式神经网络的多维度情绪回归

情绪分析是细粒度的情感分析任务,其目的是通过训练机器学习模型来判别文本中蕴含了何种情绪,是当前自然语言处理领域中的研究热点.情绪分析可细分为情绪分类与情绪回归两个任务.针对情绪回归任务,提出一种基于对抗式神经网络的多维度情绪回归方法.所提出的对抗式神经网络由3部分组成：特征抽取器、回归器、判别器.该方法旨在训练多个特征抽取器和回归器,以对输入文本的不同情绪维度进行打分.特征抽取器接受文本为输入,从文本中抽取针对不同情绪维度的特征;回归器接受由特征抽取器输出的特征为输入,对文本的不同情绪维度打分;判别器接受由特征抽取器输出的特征为输入,以判别输入的特征是针对何情绪维度.该方法借助判别器对不同的特征抽取器进行对抗式训练,从而获得能够抽取出泛化性更强的针对不同情绪维度的特征抽取器.在EMOBANK多维度情绪回归语料上的实验结果表明,该方法在EMOBANK新闻领域和小说领域的情绪回归上均取得了较为显著的性能提升,并在r值上超过了所有的基准系统,其中包括文本回归领域的先进系统.     

A comparison of decision-level sensor-fusion methods for anti-personnel landmine detection

We present the sensor-fusion results obtained from measurements within the European research project ground explosive ordinance detection (GEODE) system that strives for the realisation of a vehicle-mounted, multi-sensor, anti-personnel landmine-detection system for humanitarian de-mining. The system has three sensor types: a metal detector (MD), an infrared camera (IR), and a ground penetrating radar (GPR). The output of the sensors is processed to produce confidence levels on a grid covering the test-bed. A confidence level expresses a confidence or belief in a landmine detection on a certain position. The grid with confidence levels is the input for the decision-level sensor-fusion and provides a co-registration of the sensors. The applied fusion methods are naive Bayes' approaches, Dempster–Shafer theory, fuzzy probabilities, a rule-based method, and voting techniques. To compare fusion methods and to analyse the capacity of a method to separate landmines from the background on the basis of the output of different sensors, we provide an analysis of the different methods by viewing them as discriminant functions in the sensor confidence space. The results of experiments on real sensor data are evaluated with the leave-one-out method.     

ELM speaker identification for limited dataset using multitaper based MFCC and PNCC features with fusion score

In current scenario, speaker recognition under noisy condition is the major challenging task in the area of speech processing. Due to noise environment there is a significant degradation in the system performance. The major aim of the proposed work is to identify the speaker’s under clean and noise background using limited dataset. In this paper, we proposed a multitaper based Mel frequency cepstral coefficients (MFCC) and power normalization cepstral coefficients (PNCC) techniques with fusion strategies. Here, we used MFCC and PNCC techniques with different multitapers to extract the desired features from the obtained speech samples. Then, cepstral mean and variance normalization (CMVN) and Feature warping (FW) are the two techniques applied to normalize the obtained features from both the techniques. Furthermore, as a system model low dimension i-vector model is used and also different fusion score strategies like mean, maximum, weighted sum, cumulative and concatenated fusion techniques are utilized. Finally extreme learning machine (ELM) is used for classification in order to increase the system identification accuracy (SIA) intern which is having a single layer feedforward neural network with less complexity and time consuming compared to other neural networks. TIMIT and SITW 2016 are the two different databases are used to evaluate the proposed system under limited data of these databases. Both clean and noisy backgrounds conditions are used to check the SIA.

     

音视频双模态情感识别融合框架研究

针对双模态情感识别框架识别率低、可靠性差的问题,对情感识别最重要的两个模态语音和面部表情进行了双模态情感识别特征层融合的研究。采用基于先验知识的特征提取方法和VGGNet-19网络分别对预处理后的音视频信号进行特征提取,以直接级联的方式并通过PCA进行降维来达到特征融合的目的,使用BLSTM网络进行模型构建以完成情感识别。将该框架应用到AViD-Corpus和SEMAINE数据库上进行测试,并和传统情感识别特征层融合框架以及基于VGGNet-19或BLSTM的框架进行了对比。实验结果表明,情感识别的均方根误差（RMSE）得到降低,皮尔逊相关系数（PCC）得到提高,验证了文中提出方法的有效性。     

Driving Profile Modeling and Recognition Based on Soft Computing Approach

Advancements in biometrics-based authentication have led to its increasing prominence and are being incorporated into everyday tasks. Existing vehicle security systems rely only on alarms or smart card as forms of protection. A biometric driver recognition system utilizing driving behaviors is a highly novel and personalized approach and could be incorporated into existing vehicle security system to form a multimodal identification system and offer a greater degree of multilevel protection. In this paper, detailed studies have been conducted to model individual driving behavior in order to identify features that may be efficiently and effectively used to profile each driver. Feature extraction techniques based on Gaussian mixture models (GMMs) are proposed and implemented. Features extracted from the accelerator and brake pedal pressure were then used as inputs to a fuzzy neural network (FNN) system to ascertain the identity of the driver. Two fuzzy neural networks, namely, the evolving fuzzy neural network (EFuNN) and the adaptive network-based fuzzy inference system (ANFIS), are used to demonstrate the viability of the two proposed feature extraction techniques. The performances were compared against an artificial neural network (NN) implementation using the multilayer perceptron (MLP) network and a statistical method based on the GMM. Extensive testing was conducted and the results show great potential in the use of the FNN for real-time driver identification and verification. In addition, the profiling of driver behaviors has numerous other potential applications for use by law enforcement and companies dealing with buses and truck drivers.     

Fast transient stability assessment of large power system using probabilistic neural network with feature reduction techniques

This paper presents transient stability assessment of a large 87-bus system using a new method called the probabilistic neural network (PNN) with incorporation of feature selection and extraction methods. The investigated power system is divided into smaller areas depending on the coherency of the areas when subjected to disturbances. This is to reduce the amount of data sets collected for the respective areas. Transient stability of the power system is first determined based on the generator relative rotor angles obtained from time domain simulations carried out by considering three phase faults at different loading conditions. The data collected from the time domain simulations are then used as inputs to the PNN. Feature reduction techniques are then incorporated to reduce the number of features to the PNN which is used as a classifier to determine whether the power system is stable or unstable. It can be concluded that the PNN with the incorporation of feature reduction techniques reduces the time taken to train the PNN without affecting the accuracy of the classification results.     

多模态融合的膝关节损伤预测

膝关节磁共振成像（MRI）是诊断膝关节损伤的首选方法。然而,MRI影像的人工诊断是费时的,而且容易出现诊断错误。为了更准确地预测膝关节损伤,辅助临床医生做出诊断,提出一种多模态特征融合的深度学习模型,用于检测一般异常、前交叉韧带撕裂和半月板撕裂。提取梯度方向直方图（Histogram of Oriented Gradients, HOG）特征和局部二值模式（Local Binary Pattern,LBP）特征,经contact融合后利用PCA选取特征贡献度超过95%的特征作为传统特征;在VGG16模型的基础上加入金字塔融合的思想,将多个feature map的信息融合作为深度特征;将传统特征和深度特征经多层神经网络的能量模型进行相关性融合,作为多模态的特征,并得到预测概率。实验结果表明,上述模型在一般异常、前交叉韧带撕裂和半月板撕裂下ROC曲线下的面积（AUC）值分别为0.941?0、0.970?8和0.847?9,与传统特征和深度特征的效果相比,具有明显的优势,可以实现更有效的预测。     

轴承故障融合诊断方法

针对传统人工神经网络在故障诊断中应用的局限性 ,提出一种基于小波变换、遗传算法与神经网络的融合故障诊断方法。该方法先用小波变换对原始采样信号进行特征提取 ,再用遗传算法优化选择最为重要的特征作为神经网络的输入参数。最后 ,由神经网络进行状态识别和特征分类。这样不仅减少网络训练时间 ,降低网络计算量 ,而且有效提高分类的准确性及故障诊断的可靠性。轴承故障诊断实验结果表明 ,该方法是有效的。     

ECG beat classifier designed by combined neural network model

This paper illustrates the use of combined neural network model to guide model selection for classification of electrocardiogram (ECG) beats. The ECG signals were decomposed into time-frequency representations using discrete wavelet transform and statistical features were calculated to depict their distribution. The first level networks were implemented for ECG beats classification using the statistical features as inputs. To improve diagnostic accuracy, the second level networks were trained using the outputs of the first level networks as input data. Four types of ECG beats (normal beat, congestive heart failure beat, ventricular tachyarrhythmia beat, atrial fibrillation beat) obtained from the Physiobank database were classified with the accuracy of 96.94% by the combined neural network. The combined neural network model achieved accuracy rates which were higher than that of the stand-alone neural network model.     

An investigation of the effects of correlation and autocorrelation on classifier fusion and optimal classifier ensembles

The purpose of this research was to study various fusion strategies where the levels of correlation between features and auto-correlation within features could be controlled. The fusion strategies were chosen to reflect decision-level fusion (ISOC and ROC), feature level fusion, via a single Generalized Regression Neural Network (GRNN) employing all available features, and an intermediate level of fusion that employed the outputs of individual classifiers, in this case posterior probability estimates, before they are subjected to thresholds and mapped into decisions. This latter scheme involved fusing the posterior probability estimates by employing them as features in a probabilistic neural network. Correlation was injected into the data set both within a feature set (auto-correlation) and across feature sets, and sample size was varied for a two class problem. The fusion methods were then extended to three classifiers, and a method is demonstrated that selects the optimal classifier ensemble.     

Mask R-CNN中特征不平衡问题的全局信息融合方法

特征不平衡问题是影响神经网络检测效率的关键因素。针对Mask R-CNN中的特征不平衡问题,提出一种基于全局特征金字塔网络（GFPN）的信息融合方法。通过将GFPN产生的不同大小特征相融合,生成包含全局语义信息的特征网络,并采用反向过程对原始特征层进行重新标度,从而使得每个特征层均含有全局语义信息。实验结果表明,与原始基于Mask R-CNN的方法相比,该方法的检测精度提升4~6个百分点,而检测时间仅增加0.112 s。     

基于多模型特征与精简注意力融合的图像分类

为了提高图像分类性能,本文提出一种多模型特征和注意力模块融合的图像分类算法(image classification algorithm based on Multi-model Feature and Reduced Attention fusion,MFRA).通过多模型特征融合,使网络学习输入图像不同层次的特征,增加特征互补性,提高特征提取能力;通过加入注意力模块,使网络更关注有目标的区域,降低无关的背景干扰信息.本文算法在Cifar-10,Cifar-100,Caltech-101这3个公开数据集上的大量实验对比,验证了其有效性.与现有算法对比,本文算法的分类性能有较为明显的提升.     

Neural network and GA approaches for dwelling fire occurrence prediction

This paper describes three approaches for the prediction of dwelling fire occurrences in Derbyshire, a region in the United Kingdom. The system has been designed to calculate the number of fire occurrences for each of the 189 wards in the Derbyshire. In terms of the results from statistical analysis, eight factors are initially selected as the inputs of the neural network. Principal Component Analysis (PCA) is employed for pre-processing the input data set to reduce the number of the inputs. The first three principal components of the available data set are chosen as the inputs, the number of the fires as the output. The first approach is a logistic regression model, which has been widely used in the forest fire prediction. The prediction results of the logistic regression model are not acceptable. The second approach uses a feed-forward neural network to model the relationship between the number of fires and the factors that influence fire occurrence. The model of the neural network gives a prediction with an acceptable accuracy for the fires in dwelling areas. Genetic algorithms (GAs) are the third approach discussed in this study. The first three principle components of the available data set are classified into the different groups according to their number of fires. An iterative GA is proposed and applied to extract features for each data group. Once the features for all the groups have been identified the test data set can be easily clustered into one of the groups based on the group features. The number of fires for the group, which the test data belongs to, is the prediction of the fire occurrence for the test data. The three approaches have been compared. Our results indicate that the neural network based and the GA based approaches perform satisfactorily, with MSEs of 2.375 and 2.875, respectively, but the GA approach is much better understood and more transparent.     

Texture segmentation using pyramidal Gabor functions and self-organising feature maps

This paper presents texture segmentation realised with image treatment methods and an artificial neural network model. Gabor oriented filters are used to extract frequential texture features and Self-Organising Feature Maps are used to group and interpolate these features. In order to decrease the number of filters, we use a pyramidal multiresolution method of image representation. We intend to build an architecture inspired by the early stages of the visual cortex, while making local frequential analysis of the images, which must be able to segment different textured images.     

Object recognition using a bio-inspired neuron model with bottom-up and top-down pathways

In this paper, a new artificial neural network model is proposed for visual object recognition, in which the bottom-up, sensory-driven pathway and top-down, expectation-driven pathway are fused in information processing and their corresponding weights are learned based on the fused neuron activities. During the supervised learning process, the target labels are applied to update the bottom-up synaptic weights of the neural network. Meanwhile, the hypotheses generated by the bottom-up pathway produce expectations on sensory inputs through the top-down pathway. The expectations are constrained by the real data from the sensory inputs, which can be used to update the top-down synaptic weights accordingly. To further improve the visual object recognition performance, the multi-scale histograms of oriented gradients (MS-HOG) method is proposed to extract local features of visual objects from images. Extensive experiments on different image datasets demonstrate the efficiency and robustness of the proposed neural network model with features extracted using the MS-HOG method on visual object recognition compared with other state-of-the-art methods.     

基于U-Net融合的保留纹理的图像去噪方法

图像的噪声阻碍了高级视觉任务对图像的理解,且去除图像的噪声是一个具有挑战性的任务.现有的基于卷积神经网络的图像去噪方法在去除噪声的同时,对图像纹理会引入一定程度的破坏,导致去噪后图像无法保留图像的纹理.为了解决这个问题,本文提出一种用二分支U-Net网络来融合特征和保留纹理的图像去噪方法.首先选取一种去噪方法的两个不同去噪参数的预训练模型分别得到同一张噪声图像的不同去噪结果,其中一个结果中去噪效果比纹理保留效果好,另一个结果中纹理保留比去噪效果好.然后将这两个去噪图像作为卷积神经网络的输入,利用两个编码器分别提取图像的特征,并同时放入融合模块融合图像的特征,最后利用解码器重建出无噪声图像.实验结果表明,与现有的方法相比本文的方法更有效,在去除噪声的同时能保留更多的图像纹理信息.     

Efficient regional multi feature similarity measure based emotion detection system in web portal using artificial neural network

Emotion detection from facial expression has been well studied. There are numerous techniques has been discussed for the accuracy of emotion detection, however the methods suffer with higher false classification ratio. Towards the development of emotion detection, a novel region based multi feature similarity approach has been presented in this article. Considering, shape and geometry measure alone would not acquire higher performance in the classification. It is necessary to consider and combine multiple features towards the problem. With this motivation, the proposed Regional Multi Feature Similarity (RMFS) based emotion detection algorithm enhances the input facial image and extracts shape feature, geometry feature and wrinkle features with colors are considered. Extracted features are trained with neural network. At the classification stage, MFS measure has been estimated towards the features of various emotion class in different layers of neural network. Finally, a single one has been classified as result using artificial neural network. The proposed method improves the performance of emotion detection with reduced false ratio.     

Tissue segmentation in ultrasound images by using genetic algorithms

This paper presents a genetic based incremental neural network (GINeN) for the segmentation of tissues in ultrasound images. Performances of the GINeN and the Kohonen network are investigated for tissue segmentation in ultrasound images. Feature extraction is carried out by using continuous wavelet transform. Pixel intensities at the same spatial location on 12 wavelet planes and on the original image are considered as features, leading to 13-dimensional feature vectors. The same training set is used for the training of the Kohonen network and the GINeN.

This paper proposes the use of wavelet transform and genetic based incremental neural network together in order to increase the segmentation performance. It is observed that genetic based incremental neural network gives satisfactory segmentation performance for ultrasound images.