In Video-based Point Cloud Compression(V-PCC),2D videos to be encoded are generated by 3D point cloud projection,and compressed by High Efficiency Video Coding(HEVC).In the process of 2D video compression,the best mod...In Video-based Point Cloud Compression(V-PCC),2D videos to be encoded are generated by 3D point cloud projection,and compressed by High Efficiency Video Coding(HEVC).In the process of 2D video compression,the best mode of Coding Unit(CU)is searched by brute-force strategy,which greatly increases the complexity of the encoding process.To address this issue,we first propose a simple and effective Portable Perceptron Network(PPN)-based fast mode decision method for V-PCC under Random Access(RA)configuration.Second,we extract seven simple hand-extracted features for input into the PPN network.Third,we design an adaptive loss function,which can calculate the loss by allocating different weights according to different Rate-Distortion(RD)costs,to train our PPN network.Finally,experimental results show that the proposed method can save encoding complexity of 43.13%with almost no encoding efficiency loss under RA configuration,which is superior to the state-of-the-art methods.The source code is available at https://github.com/Mesks/PPNforV-PCC.展开更多
To improve the performance of multilayer perceptron(MLP)neural networks activated by conventional activation functions,this paper presents a new MLP activated by univariate Gaussian radial basis functions(RBFs)with ad...To improve the performance of multilayer perceptron(MLP)neural networks activated by conventional activation functions,this paper presents a new MLP activated by univariate Gaussian radial basis functions(RBFs)with adaptive centers and widths,which is composed of more than one hidden layer.In the hidden layer of the RBF-activated MLP network(MLPRBF),the outputs of the preceding layer are first linearly transformed and then fed into the univariate Gaussian RBF,which exploits the highly nonlinear property of RBF.Adaptive RBFs might address the issues of saturated outputs,low sensitivity,and vanishing gradients in MLPs activated by other prevailing nonlinear functions.Finally,we apply four MLP networks with the rectified linear unit(ReLU),sigmoid function(sigmoid),hyperbolic tangent function(tanh),and Gaussian RBF as the activation functions to approximate the one-dimensional(1D)sinusoidal function,the analytical solution of viscous Burgers’equation,and the two-dimensional(2D)steady lid-driven cavity flows.Using the same network structure,MLP-RBF generally predicts more accurately and converges faster than the other threeMLPs.MLP-RBF using less hidden layers and/or neurons per layer can yield comparable or even higher approximation accuracy than other MLPs equipped with more layers or neurons.展开更多
Owing to continuous production lines with large amount of consecutive controls, various control signals and huge logistic relations, this paper introduced the methods and principles of the development of knowledge bas...Owing to continuous production lines with large amount of consecutive controls, various control signals and huge logistic relations, this paper introduced the methods and principles of the development of knowledge base in a fault diagnosis expert system that was based on machine learning by the four-layer perceptron neural network. An example was presented. By combining differential function with not differential function and back propagation of error with back propagation of expectation, the four-layer perceptron neural network was established. And it was good for solving such a bottleneck problem in knowledge acquisition in expert system and enhancing real-time on-line diagnosis. A method of synthetic back propagation was designed, which broke the limit to non-differentiable function in BP neural network.展开更多
The attributes of the ECG signal signifying the unique electrical properties of the heart offer the opportunity to expand the realm of biometrics, which pertains the identification of an individual based on physical c...The attributes of the ECG signal signifying the unique electrical properties of the heart offer the opportunity to expand the realm of biometrics, which pertains the identification of an individual based on physical characteristics. The temporal organization of the ECG signal offers a basis for composing a machine learning feature set. The four attributes of the feature set are derived through software automation enabled by Python. These four attributes are the temporal differential of the P wave maximum and T wave maximum relative to the R wave maximum and the Q wave minimum and S wave minimum relative to the R wave maximum. The multilayer perceptron neural network was applied and evaluated in terms of classification accuracy and time to develop the model. Superior performance was achieved with respect to a reduced feature set considering only the temporal differential of the P wave maximum and T wave maximum relative to the R wave maximum by comparison to all four attributes applied to the feature set and the temporal differential of the Q wave minimum and S wave minimum relative to the R wave maximum. With these preliminary findings and the advent of portable and wearable devices for the acquisition of the ECG signal, the temporal organization of the ECG signal offers robust potential for the field of biometrics.展开更多
The Multilayer Perceptron Neural Network (MLPNN) induction technique has been successfully applied to a variety of machine learning tasks, including the extraction and classification of image features. However, not mu...The Multilayer Perceptron Neural Network (MLPNN) induction technique has been successfully applied to a variety of machine learning tasks, including the extraction and classification of image features. However, not much has been done in the application of MLPNN on images obtained by remote sensing. In this article, two automatic classification systems used in image feature extraction and classification from remote sensing data are presented. The first is a combination of two models: a MLPNN induction technique, integrated under ENVI (Environment for Visualizing Images) platform for classification, and a pre-processing model including dark subtraction for the calibration of the image, the Principal Components Analysis (PCA) for band selections and Independent Components Analysis (ICA) as blind source separator for feature extraction of the Landsat image. The second classification system is a MLPNN induction technique based on the Keras platform. In this case, there was no need for pre-processing model. Experimental results show the two classification systems to outperform other typical feature extraction and classification methods in terms of accuracy for some lithological classes including Granite1 class with the highest class accuracies of 96.69% and 92.69% for the first and second classification system respectively. Meanwhile, the two classification systems perform almost equally with the overall accuracies of 53.01% and 49.98% for the first and second models respectively </span><span style="font-family:Verdana;">though the keras model has the advantage of not integrating the pre-processing</span><span style="font-family:Verdana;"> model, hence increasing its efficiency. The application of these two systems to the study area resulted in the generation of an updated geological mapping with six lithological classes detected including the Gneiss, the Micaschist, the Schist and three versions of Granites (Granite1, Granite2 and Granite3).展开更多
This paper examines the relationship between fatal road traffic accidents and potential predictors using multilayer perceptron artificial neural network (MLANN) models. The initial analysis employed twelve potential p...This paper examines the relationship between fatal road traffic accidents and potential predictors using multilayer perceptron artificial neural network (MLANN) models. The initial analysis employed twelve potential predictors, including traffic volume, prevailing weather conditions, roadway characteristics and features, drivers’ age and gender, and number of lanes. Based on the output of the model and the variables’ importance factors, seven significant variables are identified and used for further analysis to improve the performance of models. The model is optimized by systematically changing the parameters, including the number of hidden layers and the activation function of both the hidden and output layers. The performances of the MLANN models are evaluated using the percentage of the achieved accuracy, R-squared, and Sum of Square Error (SSE) functions.展开更多
为解决现有算法在处理不同特征点云时存在普适性不足、效率低下、难以应用于工程实际的问题,提出一种基于神经网络的岩体结构面智能识别方法,具体包括4个步骤。首先,对原始点云进行标准化预处理操作,并人工选取具有代表性的特征区域,以...为解决现有算法在处理不同特征点云时存在普适性不足、效率低下、难以应用于工程实际的问题,提出一种基于神经网络的岩体结构面智能识别方法,具体包括4个步骤。首先,对原始点云进行标准化预处理操作,并人工选取具有代表性的特征区域,以构建高质量的训练样本集;其次,采用CFSFDP(clustering by fast search and find of density peaks)聚类算法为样本生成标签;再次,构建并训练多层感知机(multilayer perceptron,MLP)模型和多层卷积神经网络(multi-layer convolutional neural network,MCNN)模型,输入全尺度点云的点法向量进行结构面粗识别,并对2种模型进行比选分析;最后,使用HDBSCAN(hierarchical density-based spatial clustering of applications with noise)算法对分类结果进行细化与产状计算。结果表明:1)采用多层感知机模型处理简单结构面时具有较高的处理速度,而卷积神经网络模型在处理复杂、非均匀点云时展现出更高的分类精度。2)与聚类方法相比,该方法计算时间提升25%~50%,能够有效解决传统算法无法适用于不同复杂点云的问题,且具有很强的鲁棒性。展开更多
为进一步提高高光谱影像分类精度,通过引入Network In Network网络结构,构建了一种新的网络模型。该网络模型能够对局部感受野内的数据进行更加抽象的建模,从而能够对影像中的空谱联合特征进行更为抽象的表达。通过在Pavia University和...为进一步提高高光谱影像分类精度,通过引入Network In Network网络结构,构建了一种新的网络模型。该网络模型能够对局部感受野内的数据进行更加抽象的建模,从而能够对影像中的空谱联合特征进行更为抽象的表达。通过在Pavia University和Indian Pines两个数据集上进行验证,实验结果表明,所构建的网络模型能够有效提高分类精度,在减少训练样本的条件下仍具有较好的分类性能。展开更多
A brain-computer interface(BCI)system is one of the most effective ways that translates brain signals into output commands.Different imagery activities can be classified based on the changes inμandβrhythms and their...A brain-computer interface(BCI)system is one of the most effective ways that translates brain signals into output commands.Different imagery activities can be classified based on the changes inμandβrhythms and their spatial distributions.Multi-layer perceptron neural networks(MLP-NNs)are commonly used for classification.Training such MLP-NNs has great importance in a way that has attracted many researchers to this field recently.Conventional methods for training NNs,such as gradient descent and recursive methods,have some disadvantages including low accuracy,slow convergence speed and trapping in local minimums.In this paper,in order to overcome these issues,the MLP-NN trained by a hybrid population-physics-based algorithm,the combination of particle swarm optimization and gravitational search algorithm(PSOGSA),is proposed for our classification problem.To show the advantages of using PSOGSA that trains NNs,this algorithm is compared with other meta-heuristic algorithms such as particle swarm optimization(PSO),gravitational search algorithm(GSA)and new versions of PSO.The metrics that are discussed in this paper are the speed of convergence and classification accuracy metrics.The results show that the proposed algorithm in most subjects of encephalography(EEG)dataset has very better or acceptable performance compared to others.展开更多
基金supported by the National Natural Science Foundation of China(No.62001209).
文摘In Video-based Point Cloud Compression(V-PCC),2D videos to be encoded are generated by 3D point cloud projection,and compressed by High Efficiency Video Coding(HEVC).In the process of 2D video compression,the best mode of Coding Unit(CU)is searched by brute-force strategy,which greatly increases the complexity of the encoding process.To address this issue,we first propose a simple and effective Portable Perceptron Network(PPN)-based fast mode decision method for V-PCC under Random Access(RA)configuration.Second,we extract seven simple hand-extracted features for input into the PPN network.Third,we design an adaptive loss function,which can calculate the loss by allocating different weights according to different Rate-Distortion(RD)costs,to train our PPN network.Finally,experimental results show that the proposed method can save encoding complexity of 43.13%with almost no encoding efficiency loss under RA configuration,which is superior to the state-of-the-art methods.The source code is available at https://github.com/Mesks/PPNforV-PCC.
基金This work was partially supported by the research grant of the National University of Singapore(NUS),Ministry of Education(MOE Tier 1).
文摘To improve the performance of multilayer perceptron(MLP)neural networks activated by conventional activation functions,this paper presents a new MLP activated by univariate Gaussian radial basis functions(RBFs)with adaptive centers and widths,which is composed of more than one hidden layer.In the hidden layer of the RBF-activated MLP network(MLPRBF),the outputs of the preceding layer are first linearly transformed and then fed into the univariate Gaussian RBF,which exploits the highly nonlinear property of RBF.Adaptive RBFs might address the issues of saturated outputs,low sensitivity,and vanishing gradients in MLPs activated by other prevailing nonlinear functions.Finally,we apply four MLP networks with the rectified linear unit(ReLU),sigmoid function(sigmoid),hyperbolic tangent function(tanh),and Gaussian RBF as the activation functions to approximate the one-dimensional(1D)sinusoidal function,the analytical solution of viscous Burgers’equation,and the two-dimensional(2D)steady lid-driven cavity flows.Using the same network structure,MLP-RBF generally predicts more accurately and converges faster than the other threeMLPs.MLP-RBF using less hidden layers and/or neurons per layer can yield comparable or even higher approximation accuracy than other MLPs equipped with more layers or neurons.
文摘Owing to continuous production lines with large amount of consecutive controls, various control signals and huge logistic relations, this paper introduced the methods and principles of the development of knowledge base in a fault diagnosis expert system that was based on machine learning by the four-layer perceptron neural network. An example was presented. By combining differential function with not differential function and back propagation of error with back propagation of expectation, the four-layer perceptron neural network was established. And it was good for solving such a bottleneck problem in knowledge acquisition in expert system and enhancing real-time on-line diagnosis. A method of synthetic back propagation was designed, which broke the limit to non-differentiable function in BP neural network.
文摘The attributes of the ECG signal signifying the unique electrical properties of the heart offer the opportunity to expand the realm of biometrics, which pertains the identification of an individual based on physical characteristics. The temporal organization of the ECG signal offers a basis for composing a machine learning feature set. The four attributes of the feature set are derived through software automation enabled by Python. These four attributes are the temporal differential of the P wave maximum and T wave maximum relative to the R wave maximum and the Q wave minimum and S wave minimum relative to the R wave maximum. The multilayer perceptron neural network was applied and evaluated in terms of classification accuracy and time to develop the model. Superior performance was achieved with respect to a reduced feature set considering only the temporal differential of the P wave maximum and T wave maximum relative to the R wave maximum by comparison to all four attributes applied to the feature set and the temporal differential of the Q wave minimum and S wave minimum relative to the R wave maximum. With these preliminary findings and the advent of portable and wearable devices for the acquisition of the ECG signal, the temporal organization of the ECG signal offers robust potential for the field of biometrics.
文摘The Multilayer Perceptron Neural Network (MLPNN) induction technique has been successfully applied to a variety of machine learning tasks, including the extraction and classification of image features. However, not much has been done in the application of MLPNN on images obtained by remote sensing. In this article, two automatic classification systems used in image feature extraction and classification from remote sensing data are presented. The first is a combination of two models: a MLPNN induction technique, integrated under ENVI (Environment for Visualizing Images) platform for classification, and a pre-processing model including dark subtraction for the calibration of the image, the Principal Components Analysis (PCA) for band selections and Independent Components Analysis (ICA) as blind source separator for feature extraction of the Landsat image. The second classification system is a MLPNN induction technique based on the Keras platform. In this case, there was no need for pre-processing model. Experimental results show the two classification systems to outperform other typical feature extraction and classification methods in terms of accuracy for some lithological classes including Granite1 class with the highest class accuracies of 96.69% and 92.69% for the first and second classification system respectively. Meanwhile, the two classification systems perform almost equally with the overall accuracies of 53.01% and 49.98% for the first and second models respectively </span><span style="font-family:Verdana;">though the keras model has the advantage of not integrating the pre-processing</span><span style="font-family:Verdana;"> model, hence increasing its efficiency. The application of these two systems to the study area resulted in the generation of an updated geological mapping with six lithological classes detected including the Gneiss, the Micaschist, the Schist and three versions of Granites (Granite1, Granite2 and Granite3).
文摘This paper examines the relationship between fatal road traffic accidents and potential predictors using multilayer perceptron artificial neural network (MLANN) models. The initial analysis employed twelve potential predictors, including traffic volume, prevailing weather conditions, roadway characteristics and features, drivers’ age and gender, and number of lanes. Based on the output of the model and the variables’ importance factors, seven significant variables are identified and used for further analysis to improve the performance of models. The model is optimized by systematically changing the parameters, including the number of hidden layers and the activation function of both the hidden and output layers. The performances of the MLANN models are evaluated using the percentage of the achieved accuracy, R-squared, and Sum of Square Error (SSE) functions.
文摘为解决现有算法在处理不同特征点云时存在普适性不足、效率低下、难以应用于工程实际的问题,提出一种基于神经网络的岩体结构面智能识别方法,具体包括4个步骤。首先,对原始点云进行标准化预处理操作,并人工选取具有代表性的特征区域,以构建高质量的训练样本集;其次,采用CFSFDP(clustering by fast search and find of density peaks)聚类算法为样本生成标签;再次,构建并训练多层感知机(multilayer perceptron,MLP)模型和多层卷积神经网络(multi-layer convolutional neural network,MCNN)模型,输入全尺度点云的点法向量进行结构面粗识别,并对2种模型进行比选分析;最后,使用HDBSCAN(hierarchical density-based spatial clustering of applications with noise)算法对分类结果进行细化与产状计算。结果表明:1)采用多层感知机模型处理简单结构面时具有较高的处理速度,而卷积神经网络模型在处理复杂、非均匀点云时展现出更高的分类精度。2)与聚类方法相比,该方法计算时间提升25%~50%,能够有效解决传统算法无法适用于不同复杂点云的问题,且具有很强的鲁棒性。
文摘为进一步提高高光谱影像分类精度,通过引入Network In Network网络结构,构建了一种新的网络模型。该网络模型能够对局部感受野内的数据进行更加抽象的建模,从而能够对影像中的空谱联合特征进行更为抽象的表达。通过在Pavia University和Indian Pines两个数据集上进行验证,实验结果表明,所构建的网络模型能够有效提高分类精度,在减少训练样本的条件下仍具有较好的分类性能。
文摘A brain-computer interface(BCI)system is one of the most effective ways that translates brain signals into output commands.Different imagery activities can be classified based on the changes inμandβrhythms and their spatial distributions.Multi-layer perceptron neural networks(MLP-NNs)are commonly used for classification.Training such MLP-NNs has great importance in a way that has attracted many researchers to this field recently.Conventional methods for training NNs,such as gradient descent and recursive methods,have some disadvantages including low accuracy,slow convergence speed and trapping in local minimums.In this paper,in order to overcome these issues,the MLP-NN trained by a hybrid population-physics-based algorithm,the combination of particle swarm optimization and gravitational search algorithm(PSOGSA),is proposed for our classification problem.To show the advantages of using PSOGSA that trains NNs,this algorithm is compared with other meta-heuristic algorithms such as particle swarm optimization(PSO),gravitational search algorithm(GSA)and new versions of PSO.The metrics that are discussed in this paper are the speed of convergence and classification accuracy metrics.The results show that the proposed algorithm in most subjects of encephalography(EEG)dataset has very better or acceptable performance compared to others.
