| Convolutional neural network as the most widely used neural network model can automatically extract data features,but it is very easy to fall into the dilemma of overfitting.At the same time,the fuzzy neural network is used to describe high-order nonlinear relationships,but it cannot automatically perform feature extraction.The fuzzy polynomial neural network,as an extension of the fuzzy neural network,has the disadvantages of low system efficiency when dealing with complex and high-order problems,difficult to operate time series data,and easy to fall into the predicament of overfitting.In order to solve the above problems,we propose a convolutional fuzzy polynomial neural network.The key innovations in this paper are shown here:First,we proposed the convolutional fuzzy polynomial neural network.The innovations and superiorities of the model are divided into the following two points:(1)it can establish an association with the output under the premise of unknown input data type,(2)not limited to the excellent prediction of a small sample data set,even if the new sample set is additionally predicted,the performance is better than the traditional model.First of all,we optimized the classic fuzzy clustering algorithm with the aid of the convolution idea,and the convolutional fuzzy clustering algorithm with strong local perception is creatively proposed.At the same time,linear polynomials,as the basic description unit for dealing with unknown relationships,can also establish a correlation between input and output.Second,we proposed the convolutional fuzzy wavelet polynomial neural network.Contrasted to convolutional fuzzy neural network,the innovations and superiorities of the model are divided into the following two points(1)adding the element of wavelet function can perform calculation operations on samples in time domain space,(2)choose to add penalty items to prevent the model from learning the rules of individual abnormal data and then the entire algorithm is abnormal.First,the Morlet wavelet function is added to the construction of the convolutional fuzzy polynomial neural network neurons,which can not only broaden the range of data types processed by the model,but also resist noise attacks to a certain extent.In addition,genetic algorithms are used to optimize the model’s coding design and mutation operation.Third,a wireless gait recognition model for convolutional fuzzy polynomial neural network is proposed.It overturns the traditional video method using cameras and sensor-based identification methods.It uses commonly commercial Wi-Fi devices to achieve gait-based identification.This model uses Wi-Fi devices to collect signals,and the collected three-way wireless signal is regarded as the three color components of image data for feature extraction.The human gait recognition application of wireless signals is realized,and the model is tested and classified in real scenes with accurate and efficient results. |
