第五代(fifth-generation,5G)移动通信技术的兴起,推动了物联网(Internet of things,IoT)的发展。然而,随着物联网数据传输量的爆发式增长,频谱资源短缺问题越来越严重。频谱感知技术极大的提高了物联网频谱利用率。但是,物联网移动通...第五代(fifth-generation,5G)移动通信技术的兴起,推动了物联网(Internet of things,IoT)的发展。然而,随着物联网数据传输量的爆发式增长,频谱资源短缺问题越来越严重。频谱感知技术极大的提高了物联网频谱利用率。但是,物联网移动通信环境的复杂性高以及信号易畸变的特性,对现有的频谱感知算法提出了重大挑战。因此,提出了一种融合去噪自编码器(denoising autoencoder,DAE)和改进长短时记忆(long short term memory,LSTM)神经网络的智能频谱感知算法。DAE通过编码和解码过程挖掘移动信号的底层结构特征,改进的LSTM频谱感知分类器模型结合过去时刻信息特征对时序信号序列进行分类。与支持向量机(support vector machine,SVM)、循环神经网络(recurrent neural network,RNN)、LeNet5、学习矢量量化(learning vector quantization,LVQ)和Elman算法相比,该算法的感知性能提高了45%。展开更多
Time series analysis is a key technology for medical diagnosis,weather forecasting and financial prediction systems.However,missing data frequently occur during data recording,posing a great challenge to data mining t...Time series analysis is a key technology for medical diagnosis,weather forecasting and financial prediction systems.However,missing data frequently occur during data recording,posing a great challenge to data mining tasks.In this study,we propose a novel time series data representation-based denoising autoencoder(DAE)for the reconstruction of missing values.Two data representation methods,namely,recurrence plot(RP)and Gramian angular field(GAF),are used to transform the raw time series to a 2D matrix for establishing the temporal correlations between different time intervals and extracting the structural patterns from the time series.Then an improved DAE is proposed to reconstruct the missing values from the 2D representation of time series.A comprehensive comparison is conducted amongst the different representations on standard datasets.Results show that the 2D representations have a lower reconstruction error than the raw time series,and the RP representation provides the best outcome.This work provides useful insights into the better reconstruction of missing values in time series analysis to considerably improve the reliability of timevarying system.展开更多
Offline Urdu Nastaleeq text recognition has long been a serious problem due to its very cursive nature. In order to get rid of the character segmentation problems, many researchers are shifting focus towards segmentat...Offline Urdu Nastaleeq text recognition has long been a serious problem due to its very cursive nature. In order to get rid of the character segmentation problems, many researchers are shifting focus towards segmentation free ligature based recognition approaches. Majority of the prevalent ligature based recognition systems heavily rely on hand-engineered feature extraction techniques. However, such techniques are more error prone and may often lead to a loss of useful information that might hardly be captured later by any manual features. Most of the prevalent Urdu Nastaleeq test recognition was trained and tested on small sets. This paper proposes the use of stacked denoising autoencoder for automatic feature extraction directly from raw pixel values of ligature images. Such deep learning networks have not been applied for the recognition of Urdu text thus far. Different stacked denoising autoencoders have been trained on 178573 ligatures with 3732 classes from un-degraded(noise free) UPTI(Urdu Printed Text Image) data set. Subsequently, trained networks are validated and tested on degraded versions of UPTI data set. The experimental results demonstrate accuracies in range of 93% to 96% which are better than the existing Urdu OCR systems for such large dataset of ligatures.展开更多
Unmanned surface vehicle(USV)is currently a hot research topic in maritime communication network(MCN),where denoising and semantic segmentation of maritime images taken by USV have been rarely studied.The former has r...Unmanned surface vehicle(USV)is currently a hot research topic in maritime communication network(MCN),where denoising and semantic segmentation of maritime images taken by USV have been rarely studied.The former has recently researched on autoencoder model used for image denoising,but the existed models are too complicated to be suitable for real-time detection of USV.In this paper,we proposed a lightweight autoencoder combined with inception module for maritime image denoising in different noisy environments and explore the effect of different inception modules on the denoising performance.Furthermore,we completed the semantic segmentation task for maritime images taken by USV utilizing the pretrained U-Net model with tuning,and compared them with original U-Net model based on different backbone.Subsequently,we compared the semantic segmentation of noised and denoised maritime images respectively to explore the effect of image noise on semantic segmentation performance.Case studies are provided to prove the feasibility of our proposed denoising and segmentation method.Finally,a simple integrated communication system combining image denoising and segmentation for USV is shown.展开更多
We study the effects of quantization and additive white Gaussian noise(AWGN) in transmitting latent representations of images over a noisy communication channel. The latent representations are obtained using autoencod...We study the effects of quantization and additive white Gaussian noise(AWGN) in transmitting latent representations of images over a noisy communication channel. The latent representations are obtained using autoencoders(AEs). We analyze image reconstruction and classification performance for different channel noise powers, latent vector sizes, and number of quantization bits used for the latent variables as well as AEs’ parameters. The results show that the digital transmission of latent representations using conventional AEs alone is extremely vulnerable to channel noise and quantization effects. We then propose a combination of basic AE and a denoising autoencoder(DAE) to denoise the corrupted latent vectors at the receiver. This approach demonstrates robustness against channel noise and quantization effects and enables a significant improvement in image reconstruction and classification performance particularly in adverse scenarios with high noise powers and significant quantization effects.展开更多
流量异常检测能够有效识别网络流量数据中的攻击行为,是一种重要的网络安全防护手段。近年来,深度学习在流量异常检测领域得到了广泛应用,现有的深度学习模型进行流量异常检测存在两个问题:一是数据受噪声影响导致检测鲁棒性差、准确率...流量异常检测能够有效识别网络流量数据中的攻击行为,是一种重要的网络安全防护手段。近年来,深度学习在流量异常检测领域得到了广泛应用,现有的深度学习模型进行流量异常检测存在两个问题:一是数据受噪声影响导致检测鲁棒性差、准确率低;二是数据特征维度高以及模型参数多导致训练和检测速度慢。为了在降低流量数据噪声影响的基础上提高检测速度和准确性,本文提出了一种基于去噪自编码器(Denoising Auto Encoder, DAE)和门控循环单元(Gated Recurrent Unit, GRU)组合的流量异常检测方法。首先设计了基于DAE的流量特征提取算法,采用小批量梯度下降算法对DAE进行训练,通过最小化含噪声数据的重构向量与原始输入向量间的差异,有效提取具有较强鲁棒性的流量特征,降低特征维度。然后设计了基于GRU的异常检测算法,利用提取的低维流量特征数据训练GRU,从而构建异常流量分类器,实现对攻击流量的准确检测。最后在NSL-KDD、UNSW-NB15、CICIDS2017数据集上的实验结果表明:与其他的机器学习、深度学习方法相比,本文所提方法的检测准确率最大提升了18.71%。同时,本文方法可以实现较高的精确率、召回率和检测效率,同时具有较低的误报率。在面对数据受到噪声破坏时,具有较强的检测鲁棒性。展开更多
Wireless sensor networks are increasingly used in sensitive event monitoring.However,various abnormal data generated by sensors greatly decrease the accuracy of the event detection.Although many methods have been prop...Wireless sensor networks are increasingly used in sensitive event monitoring.However,various abnormal data generated by sensors greatly decrease the accuracy of the event detection.Although many methods have been proposed to deal with the abnormal data,they generally detect and/or repair all abnormal data without further differentiate.Actually,besides the abnormal data caused by events,it is well known that sensor nodes prone to generate abnormal data due to factors such as sensor hardware drawbacks and random effects of external sources.Dealing with all abnormal data without differentiate will result in false detection or missed detection of the events.In this paper,we propose a data cleaning approach based on Stacked Denoising Autoencoders(SDAE)and multi-sensor collaborations.We detect all abnormal data by SDAE,then differentiate the abnormal data by multi-sensor collaborations.The abnormal data caused by events are unchanged,while the abnormal data caused by other factors are repaired.Real data based simulations show the efficiency of the proposed approach.展开更多
Extreme learning machine (ELM) is a feedforward neural network-based machine learning method that has the benefits of short training times, strong generalization capabilities, and will not fall into local minima. Howe...Extreme learning machine (ELM) is a feedforward neural network-based machine learning method that has the benefits of short training times, strong generalization capabilities, and will not fall into local minima. However, due to the traditional ELM shallow architecture, it requires a large number of hidden nodes when dealing with high-dimensional data sets to ensure its classification performance. The other aspect, it is easy to degrade the classification performance in the face of noise interference from noisy data. To improve the above problem, this paper proposes a double pseudo-inverse extreme learning machine (DPELM) based on Sparse Denoising AutoEncoder (SDAE) namely, SDAE-DPELM. The algorithm can directly determine the input weight and output weight of the network by using the pseudo-inverse method. As a result, the algorithm only requires a few hidden layer nodes to produce superior classification results when classifying data. And its combination with SDAE can effectively improve the classification performance and noise resistance. Extensive numerical experiments show that the algorithm has high classification accuracy and good robustness when dealing with high-dimensional noisy data and high-dimensional noiseless data. Furthermore, applying such an algorithm to Miao character recognition substantiates its excellent performance, which further illustrates the practicability of the algorithm.展开更多
Invoice document digitization is crucial for efficient management in industries.The scanned invoice image is often noisy due to various reasons.This affects the OCR(optical character recognition)detection accuracy.In ...Invoice document digitization is crucial for efficient management in industries.The scanned invoice image is often noisy due to various reasons.This affects the OCR(optical character recognition)detection accuracy.In this paper,letter data obtained from images of invoices are denoised using a modified autoencoder based deep learning method.A stacked denoising autoencoder(SDAE)is implemented with two hidden layers each in encoder network and decoder network.In order to capture the most salient features of training samples,a undercomplete autoencoder is designed with non-linear encoder and decoder function.This autoencoder is regularized for denoising application using a combined loss function which considers both mean square error and binary cross entropy.A dataset consisting of 59,119 letter images,which contains both English alphabets(upper and lower case)and numbers(0 to 9)is prepared from many scanned invoices images and windows true type(.ttf)files,are used for training the neural network.Performance is analyzed in terms of Signal to Noise Ratio(SNR),Peak Signal to Noise Ratio(PSNR),Structural Similarity Index(SSIM)and Universal Image Quality Index(UQI)and compared with other filtering techniques like Nonlocal Means filter,Anisotropic diffusion filter,Gaussian filters and Mean filters.Denoising performance of proposed SDAE is compared with existing SDAE with single loss function in terms of SNR and PSNR values.Results show the superior performance of proposed SDAE method.展开更多
Object detection,one of the core research topics in computer vision,is extensively used in various industrial activities.Although there have been many studies of daytime images where objects can be easily detected,the...Object detection,one of the core research topics in computer vision,is extensively used in various industrial activities.Although there have been many studies of daytime images where objects can be easily detected,there is relatively little research on nighttime images.In the case of nighttime,various types of noises,such as darkness,haze,and light blur,deteriorate image quality.Thus,an appropriate process for removing noise must precede to improve object detection performance.Although there are many studies on removing individual noise,only a few studies handle multiple noises simultaneously.In this paper,we pro-pose a convolutional denoising autoencoder(CDAE)-based architecture trained on various types of noises.We also present various composing modules for each noise to improve object detection performance for night images.Using the exclusively dark(ExDark)Image dataset,experimental results show that the Sequentialfiltering architecture showed superior mean average precision(mAP)compared to other architectures.展开更多
文摘第五代(fifth-generation,5G)移动通信技术的兴起,推动了物联网(Internet of things,IoT)的发展。然而,随着物联网数据传输量的爆发式增长,频谱资源短缺问题越来越严重。频谱感知技术极大的提高了物联网频谱利用率。但是,物联网移动通信环境的复杂性高以及信号易畸变的特性,对现有的频谱感知算法提出了重大挑战。因此,提出了一种融合去噪自编码器(denoising autoencoder,DAE)和改进长短时记忆(long short term memory,LSTM)神经网络的智能频谱感知算法。DAE通过编码和解码过程挖掘移动信号的底层结构特征,改进的LSTM频谱感知分类器模型结合过去时刻信息特征对时序信号序列进行分类。与支持向量机(support vector machine,SVM)、循环神经网络(recurrent neural network,RNN)、LeNet5、学习矢量量化(learning vector quantization,LVQ)和Elman算法相比,该算法的感知性能提高了45%。
文摘Time series analysis is a key technology for medical diagnosis,weather forecasting and financial prediction systems.However,missing data frequently occur during data recording,posing a great challenge to data mining tasks.In this study,we propose a novel time series data representation-based denoising autoencoder(DAE)for the reconstruction of missing values.Two data representation methods,namely,recurrence plot(RP)and Gramian angular field(GAF),are used to transform the raw time series to a 2D matrix for establishing the temporal correlations between different time intervals and extracting the structural patterns from the time series.Then an improved DAE is proposed to reconstruct the missing values from the 2D representation of time series.A comprehensive comparison is conducted amongst the different representations on standard datasets.Results show that the 2D representations have a lower reconstruction error than the raw time series,and the RP representation provides the best outcome.This work provides useful insights into the better reconstruction of missing values in time series analysis to considerably improve the reliability of timevarying system.
基金National Natural Science Foundation of China (Project No. 61273365)111 Project (No. B08004) are gratefully acknowledged
文摘Offline Urdu Nastaleeq text recognition has long been a serious problem due to its very cursive nature. In order to get rid of the character segmentation problems, many researchers are shifting focus towards segmentation free ligature based recognition approaches. Majority of the prevalent ligature based recognition systems heavily rely on hand-engineered feature extraction techniques. However, such techniques are more error prone and may often lead to a loss of useful information that might hardly be captured later by any manual features. Most of the prevalent Urdu Nastaleeq test recognition was trained and tested on small sets. This paper proposes the use of stacked denoising autoencoder for automatic feature extraction directly from raw pixel values of ligature images. Such deep learning networks have not been applied for the recognition of Urdu text thus far. Different stacked denoising autoencoders have been trained on 178573 ligatures with 3732 classes from un-degraded(noise free) UPTI(Urdu Printed Text Image) data set. Subsequently, trained networks are validated and tested on degraded versions of UPTI data set. The experimental results demonstrate accuracies in range of 93% to 96% which are better than the existing Urdu OCR systems for such large dataset of ligatures.
基金Natural Science Foundation of Fujian Province(No.2019J05026)in part by the Education Scientific Research Project for Young Teachers of Fujian Province(No.JT180053).
文摘Unmanned surface vehicle(USV)is currently a hot research topic in maritime communication network(MCN),where denoising and semantic segmentation of maritime images taken by USV have been rarely studied.The former has recently researched on autoencoder model used for image denoising,but the existed models are too complicated to be suitable for real-time detection of USV.In this paper,we proposed a lightweight autoencoder combined with inception module for maritime image denoising in different noisy environments and explore the effect of different inception modules on the denoising performance.Furthermore,we completed the semantic segmentation task for maritime images taken by USV utilizing the pretrained U-Net model with tuning,and compared them with original U-Net model based on different backbone.Subsequently,we compared the semantic segmentation of noised and denoised maritime images respectively to explore the effect of image noise on semantic segmentation performance.Case studies are provided to prove the feasibility of our proposed denoising and segmentation method.Finally,a simple integrated communication system combining image denoising and segmentation for USV is shown.
基金supported by Hong Kong Government general research fund (GRF) under project number PolyU152757/16ENational Natural Science Foundation China under project numbers 61435006 and 61401020
文摘We study the effects of quantization and additive white Gaussian noise(AWGN) in transmitting latent representations of images over a noisy communication channel. The latent representations are obtained using autoencoders(AEs). We analyze image reconstruction and classification performance for different channel noise powers, latent vector sizes, and number of quantization bits used for the latent variables as well as AEs’ parameters. The results show that the digital transmission of latent representations using conventional AEs alone is extremely vulnerable to channel noise and quantization effects. We then propose a combination of basic AE and a denoising autoencoder(DAE) to denoise the corrupted latent vectors at the receiver. This approach demonstrates robustness against channel noise and quantization effects and enables a significant improvement in image reconstruction and classification performance particularly in adverse scenarios with high noise powers and significant quantization effects.
文摘流量异常检测能够有效识别网络流量数据中的攻击行为,是一种重要的网络安全防护手段。近年来,深度学习在流量异常检测领域得到了广泛应用,现有的深度学习模型进行流量异常检测存在两个问题:一是数据受噪声影响导致检测鲁棒性差、准确率低;二是数据特征维度高以及模型参数多导致训练和检测速度慢。为了在降低流量数据噪声影响的基础上提高检测速度和准确性,本文提出了一种基于去噪自编码器(Denoising Auto Encoder, DAE)和门控循环单元(Gated Recurrent Unit, GRU)组合的流量异常检测方法。首先设计了基于DAE的流量特征提取算法,采用小批量梯度下降算法对DAE进行训练,通过最小化含噪声数据的重构向量与原始输入向量间的差异,有效提取具有较强鲁棒性的流量特征,降低特征维度。然后设计了基于GRU的异常检测算法,利用提取的低维流量特征数据训练GRU,从而构建异常流量分类器,实现对攻击流量的准确检测。最后在NSL-KDD、UNSW-NB15、CICIDS2017数据集上的实验结果表明:与其他的机器学习、深度学习方法相比,本文所提方法的检测准确率最大提升了18.71%。同时,本文方法可以实现较高的精确率、召回率和检测效率,同时具有较低的误报率。在面对数据受到噪声破坏时,具有较强的检测鲁棒性。
基金This work is supported by the National Natural Science Foundation of China(Grant No.61672282)the Basic Research Program of Jiangsu Province(Grant No.BK20161491).
文摘Wireless sensor networks are increasingly used in sensitive event monitoring.However,various abnormal data generated by sensors greatly decrease the accuracy of the event detection.Although many methods have been proposed to deal with the abnormal data,they generally detect and/or repair all abnormal data without further differentiate.Actually,besides the abnormal data caused by events,it is well known that sensor nodes prone to generate abnormal data due to factors such as sensor hardware drawbacks and random effects of external sources.Dealing with all abnormal data without differentiate will result in false detection or missed detection of the events.In this paper,we propose a data cleaning approach based on Stacked Denoising Autoencoders(SDAE)and multi-sensor collaborations.We detect all abnormal data by SDAE,then differentiate the abnormal data by multi-sensor collaborations.The abnormal data caused by events are unchanged,while the abnormal data caused by other factors are repaired.Real data based simulations show the efficiency of the proposed approach.
文摘Extreme learning machine (ELM) is a feedforward neural network-based machine learning method that has the benefits of short training times, strong generalization capabilities, and will not fall into local minima. However, due to the traditional ELM shallow architecture, it requires a large number of hidden nodes when dealing with high-dimensional data sets to ensure its classification performance. The other aspect, it is easy to degrade the classification performance in the face of noise interference from noisy data. To improve the above problem, this paper proposes a double pseudo-inverse extreme learning machine (DPELM) based on Sparse Denoising AutoEncoder (SDAE) namely, SDAE-DPELM. The algorithm can directly determine the input weight and output weight of the network by using the pseudo-inverse method. As a result, the algorithm only requires a few hidden layer nodes to produce superior classification results when classifying data. And its combination with SDAE can effectively improve the classification performance and noise resistance. Extensive numerical experiments show that the algorithm has high classification accuracy and good robustness when dealing with high-dimensional noisy data and high-dimensional noiseless data. Furthermore, applying such an algorithm to Miao character recognition substantiates its excellent performance, which further illustrates the practicability of the algorithm.
文摘Invoice document digitization is crucial for efficient management in industries.The scanned invoice image is often noisy due to various reasons.This affects the OCR(optical character recognition)detection accuracy.In this paper,letter data obtained from images of invoices are denoised using a modified autoencoder based deep learning method.A stacked denoising autoencoder(SDAE)is implemented with two hidden layers each in encoder network and decoder network.In order to capture the most salient features of training samples,a undercomplete autoencoder is designed with non-linear encoder and decoder function.This autoencoder is regularized for denoising application using a combined loss function which considers both mean square error and binary cross entropy.A dataset consisting of 59,119 letter images,which contains both English alphabets(upper and lower case)and numbers(0 to 9)is prepared from many scanned invoices images and windows true type(.ttf)files,are used for training the neural network.Performance is analyzed in terms of Signal to Noise Ratio(SNR),Peak Signal to Noise Ratio(PSNR),Structural Similarity Index(SSIM)and Universal Image Quality Index(UQI)and compared with other filtering techniques like Nonlocal Means filter,Anisotropic diffusion filter,Gaussian filters and Mean filters.Denoising performance of proposed SDAE is compared with existing SDAE with single loss function in terms of SNR and PSNR values.Results show the superior performance of proposed SDAE method.
基金supported by the Ministry of Education of the Republic of Korea and the National Research Foundation of Korea(NRF-2021S1A5A2A01061459).
文摘Object detection,one of the core research topics in computer vision,is extensively used in various industrial activities.Although there have been many studies of daytime images where objects can be easily detected,there is relatively little research on nighttime images.In the case of nighttime,various types of noises,such as darkness,haze,and light blur,deteriorate image quality.Thus,an appropriate process for removing noise must precede to improve object detection performance.Although there are many studies on removing individual noise,only a few studies handle multiple noises simultaneously.In this paper,we pro-pose a convolutional denoising autoencoder(CDAE)-based architecture trained on various types of noises.We also present various composing modules for each noise to improve object detection performance for night images.Using the exclusively dark(ExDark)Image dataset,experimental results show that the Sequentialfiltering architecture showed superior mean average precision(mAP)compared to other architectures.