Due to the influence of terrain structure,meteorological conditions and various factors,there are anomalous data in automatic dependent surveillance-broadcast(ADS-B)message.The ADS-B equipment can be used for position...Due to the influence of terrain structure,meteorological conditions and various factors,there are anomalous data in automatic dependent surveillance-broadcast(ADS-B)message.The ADS-B equipment can be used for positioning of general aviation aircraft.Aim to acquire the accurate position information of aircraft and detect anomaly data,the ADS-B anomaly data detection model based on deep learning and difference of Gaussian(DoG)approach is proposed.First,according to the characteristic of ADS-B data,the ADS-B position data are transformed into the coordinate system.And the origin of the coordinate system is set up as the take-off point.Then,based on the kinematic principle,the ADS-B anomaly data can be removed.Moreover,the details of the ADS-B position data can be got by the DoG approach.Finally,the long short-term memory(LSTM)neural network is used to optimize the recurrent neural network(RNN)with severe gradient reduction for processing ADS-B data.The position data of ADS-B are reconstructed by the sequence to sequence(seq2seq)model which is composed of LSTM neural network,and the reconstruction error is used to detect the anomalous data.Based on the real flight data of general aviation aircraft,the simulation results show that the anomaly data can be detected effectively by the proposed method of reconstructing ADS-B data with the seq2seq model,and its running time is reduced.Compared with the RNN,the accuracy of anomaly detection is increased by 2.7%.The performance of the proposed model is better than that of the traditional anomaly detection models.展开更多
Human Action Recognition(HAR)attempts to recognize the human action from images and videos.The major challenge in HAR is the design of an action descriptor that makes the HAR system robust for different environments.A...Human Action Recognition(HAR)attempts to recognize the human action from images and videos.The major challenge in HAR is the design of an action descriptor that makes the HAR system robust for different environments.A novel action descriptor is proposed in this study,based on two independent spatial and spectral filters.The proposed descriptor uses a Difference of Gaussian(DoG)filter to extract scale-invariant features and a Difference of Wavelet(DoW)filter to extract spectral information.To create a composite feature vector for a particular test action picture,the Discriminant of Guassian(DoG)and Difference of Wavelet(DoW)features are combined.Linear Discriminant Analysis(LDA),a widely used dimensionality reduction technique,is also used to eliminate duplicate data.Finally,a closest neighbor method is used to classify the dataset.Weizmann and UCF 11 datasets were used to run extensive simulations of the suggested strategy,and the accuracy assessed after the simulations were run on Weizmann datasets for five-fold cross validation is shown to perform well.The average accuracy of DoG+DoW is observed as 83.6635%while the average accuracy of Discrinanat of Guassian(DoG)and Difference of Wavelet(DoW)is observed as 80.2312%and 77.4215%,respectively.The average accuracy measured after the simulation of proposed methods over UCF 11 action dataset for five-fold cross validation DoG+DoW is observed as 62.5231%while the average accuracy of Difference of Guassian(DoG)and Difference of Wavelet(DoW)is observed as 60.3214%and 58.1247%,respectively.From the above accuracy observations,the accuracy of Weizmann is high compared to the accuracy of UCF 11,hence verifying the effectiveness in the improvisation of recognition accuracy.展开更多
Human information processing depends mainly on billions of neurons which constitute a complex neural network,and the information is transmitted in the form of neural spikes.In this paper,we propose a spiking neural ne...Human information processing depends mainly on billions of neurons which constitute a complex neural network,and the information is transmitted in the form of neural spikes.In this paper,we propose a spiking neural network(SNN),named MD-SNN,with three key features:(1) using receptive field to encode spike trains from images;(2) randomly selecting partial spikes as inputs for each neuron to approach the absolute refractory period of the neuron;(3) using groups of neurons to make decisions.We test MD-SNN on the MNIST data set of handwritten digits,and results demonstrate that:(1) Different sizes of receptive fields influence classification results significantly.(2) Considering the neuronal refractory period in the SNN model,increasing the number of neurons in the learning layer could greatly reduce the training time,effectively reduce the probability of over-fitting,and improve the accuracy by 8.77%.(3) Compared with other SNN methods,MD-SNN achieves a better classification;compared with the convolution neural network,MD-SNN maintains flip and rotation invariance(the accuracy can remain at 90.44% on the test set),and it is more suitable for small sample learning(the accuracy can reach 80.15%for 1000 training samples,which is 7.8 times that of CNN).展开更多
基金supported by the National Key R&D Program of China(No.2018AAA0100804)the Talent Project of Revitalization Liaoning(No.XLYC1907022)+5 种基金the Key R&D Projects of Liaoning Province(No.2020JH2/10100045)the Capacity Building of Civil Aviation Safety(No.TMSA1614)the Natural Science Foundation of Liaoning Province(No.2019-MS-251)the Scientific Research Project of Liaoning Provincial Department of Education(Nos.L201705,L201716)the High-Level Innovation Talent Project of Shenyang(No.RC190030)the Second Young and Middle-Aged Talents Support Program of Shenyang Aerospace University.
文摘Due to the influence of terrain structure,meteorological conditions and various factors,there are anomalous data in automatic dependent surveillance-broadcast(ADS-B)message.The ADS-B equipment can be used for positioning of general aviation aircraft.Aim to acquire the accurate position information of aircraft and detect anomaly data,the ADS-B anomaly data detection model based on deep learning and difference of Gaussian(DoG)approach is proposed.First,according to the characteristic of ADS-B data,the ADS-B position data are transformed into the coordinate system.And the origin of the coordinate system is set up as the take-off point.Then,based on the kinematic principle,the ADS-B anomaly data can be removed.Moreover,the details of the ADS-B position data can be got by the DoG approach.Finally,the long short-term memory(LSTM)neural network is used to optimize the recurrent neural network(RNN)with severe gradient reduction for processing ADS-B data.The position data of ADS-B are reconstructed by the sequence to sequence(seq2seq)model which is composed of LSTM neural network,and the reconstruction error is used to detect the anomalous data.Based on the real flight data of general aviation aircraft,the simulation results show that the anomaly data can be detected effectively by the proposed method of reconstructing ADS-B data with the seq2seq model,and its running time is reduced.Compared with the RNN,the accuracy of anomaly detection is increased by 2.7%.The performance of the proposed model is better than that of the traditional anomaly detection models.
文摘Human Action Recognition(HAR)attempts to recognize the human action from images and videos.The major challenge in HAR is the design of an action descriptor that makes the HAR system robust for different environments.A novel action descriptor is proposed in this study,based on two independent spatial and spectral filters.The proposed descriptor uses a Difference of Gaussian(DoG)filter to extract scale-invariant features and a Difference of Wavelet(DoW)filter to extract spectral information.To create a composite feature vector for a particular test action picture,the Discriminant of Guassian(DoG)and Difference of Wavelet(DoW)features are combined.Linear Discriminant Analysis(LDA),a widely used dimensionality reduction technique,is also used to eliminate duplicate data.Finally,a closest neighbor method is used to classify the dataset.Weizmann and UCF 11 datasets were used to run extensive simulations of the suggested strategy,and the accuracy assessed after the simulations were run on Weizmann datasets for five-fold cross validation is shown to perform well.The average accuracy of DoG+DoW is observed as 83.6635%while the average accuracy of Discrinanat of Guassian(DoG)and Difference of Wavelet(DoW)is observed as 80.2312%and 77.4215%,respectively.The average accuracy measured after the simulation of proposed methods over UCF 11 action dataset for five-fold cross validation DoG+DoW is observed as 62.5231%while the average accuracy of Difference of Guassian(DoG)and Difference of Wavelet(DoW)is observed as 60.3214%and 58.1247%,respectively.From the above accuracy observations,the accuracy of Weizmann is high compared to the accuracy of UCF 11,hence verifying the effectiveness in the improvisation of recognition accuracy.
基金supported by the National Natural Science Foundation of China(Nos.61773312,61773307,and L1522023)the China Postdoctoral Science Foundation(No.2016M590949)the National Basic Research Program(973)of China(No.2015CB351703)
文摘Human information processing depends mainly on billions of neurons which constitute a complex neural network,and the information is transmitted in the form of neural spikes.In this paper,we propose a spiking neural network(SNN),named MD-SNN,with three key features:(1) using receptive field to encode spike trains from images;(2) randomly selecting partial spikes as inputs for each neuron to approach the absolute refractory period of the neuron;(3) using groups of neurons to make decisions.We test MD-SNN on the MNIST data set of handwritten digits,and results demonstrate that:(1) Different sizes of receptive fields influence classification results significantly.(2) Considering the neuronal refractory period in the SNN model,increasing the number of neurons in the learning layer could greatly reduce the training time,effectively reduce the probability of over-fitting,and improve the accuracy by 8.77%.(3) Compared with other SNN methods,MD-SNN achieves a better classification;compared with the convolution neural network,MD-SNN maintains flip and rotation invariance(the accuracy can remain at 90.44% on the test set),and it is more suitable for small sample learning(the accuracy can reach 80.15%for 1000 training samples,which is 7.8 times that of CNN).
