Research On Information Perception Technology Based On Deep Learning And Image Processing

The purpose of information perception is to obtain information that users are interested in,and to provide an important source of information for users in various applications.The most basic function of information perception is to collect a large amount of data.In order to analyze and extract hidden important information from a large amount of data,deep learning algorithms are often used as an effective tool.As an important carrier of information,images can express the information characteristics of data more intuitively than ordinary data.With the development of image technology,images are often used for target identification and information extraction.Based on the large amount of geomagnetic sensor data collected,this paper uses image conversion and deep learning to realize the automatic perception of vehicle information.According to the large amount of communication signal data generated,this paper uses image processing to complete the perception of frequency hopping signals,which is mainly divided into two parts: frequency hopping signal detection and frequency hopping signal parameter estimation.Finally,on the basis of the various types of modulation data generated,deep learning is used to achieve the purpose of automatic perception of modulation information.The main research completed in this paper are:(1)Geomagnetic vehicle perception: the geomagnetic vehicle perception studied in this paper uses geomagnetic data to classify vehicle types.Based on the magnetic field data collected by a single low-cost magnetic sensor,a convolutional neural network(CNN)model is designed to achieve vehicle classification.Specifically,first,a vehicle flow detection algorithm based on a dynamic threshold is designed,and a vehicle detection state machine is designed on the basis of the algorithm,which separates the magnetic field data of each vehicle from a large amount of collected magnetic field data.The collected data is then converted into two-dimensional grayscale images,so that the magnetic field signal images of different types of vehicles can be used as input data for CNN.Finally,by designing a CNN model suitable for vehicle model classification,it is used as a classifier for vehicle types information perception.Using the designed CNN model,this paper divides the vehicles into 7 types.The performance of the proposed vehicle classification scheme is evaluated through experiments.The experimental results show that the accuracy of vehicle classification is as high as 97.83%.(2)Frequency hopping signal perception: the frequency hopping signal perception studied in this paper refers to the detection and parameter estimation of frequency hopping signals in a complex environment.After the frequency hopping signal undergoes short-time Fourier transform(STFT),its time-frequency diagram can be used as a digital image with special content.First,by converting the time-frequency image into a simple binary image,the difficulty of image processing is reduced,and the characteristics of the frequency hopping signal are also enhanced.Then,according to the energy characteristics of the noise and the signal,the background noise is removed by the method based on the point value statistics of the time-frequency matrix.Then use appropriate structural elements to apply morphological closing and opening cascade operations to eliminate fixed-frequency interference;then based on the method of connected region marking to eliminate sweeping interference.And for the problem of residual frequency sweep interference,this article continues to use morphological operations to eliminate residual frequency sweep signals.After the frequency hopping signal is detected,the connected area of the signal is marked again,and the take-off time,frequency hopping rate and frequency set of the frequency hopping signal are estimated based on the circumscribed rectangle parameters used to mark the signal.Simulation results and analysis show that the methods have a good performance.