Frequency Hopping Signal Analysis And Recognition Technology Of Civil UAV

Unmanned aerial vehicle(UAV)originated in the military field,and quickly became popularity in the civilian field due to their miniaturization and diversity of functions.The safety of civil UAV caused by its widening application has attracted increasing public concern.Meanwhile,the signal identification of UAV is not only an advance and important link of statutory schemes,but also is the hot spot in technology and products,which is of great significance and commercial value in the anti-UAV field.This thesis analyzes the characteristics of UAV communication signal and designs two methods of UAV signal identification by image processing,realizing frequency hopping signal identification under complex electromagnetic environment.The main innovation and work are as follows.1.Analyze the characteristics of system structure and communication signal of civil UAV,collect the UAV signals,and observe the coarse-grained of whole UAV remote signal;summarize the structure and characteristics of frequency hopping communication system and simulate this system using Simulink toolkit;introduce the time-frequency analysis method for processing non-stationary signals and compare the performances and features of various methods through simulation experiments.2.The existing methods of frequency hopping signal detection and recognition based on time-frequency characteristics have the problem of insufficient recognition ability when they depend on manual threshold setting and mixing multiple interference signals.In this thesis,a method based on connected region labeling is designed to detect frequency hopping signal and estimate its parameters in complex electromagnetic environment.Firstly,the acquired signals are converted in time-frequency domain,and the bottom noise is removed by self-adapting de-noising threshold based on energy statistics.After binarization and morphological processing,the time-frequency data are labeled and the region information is extracted.The interference signal can be further processed by modifying the connected region,and the frequency-hopping signal can be sorted by clustering the connected region.Finally,according to the sorting results,the frequency hopping signal can be detected and the parameters of the target frequency hopping signal can be counted.The algorithm has self-adaptability in the process of removing white noise,which makes up for the shortage of manual intervention in setting denoising threshold.In the process of frequency hopping signal detection,there is no need to know the prior condition of target signal hopping,and the method of dealing with skewed interference is given.The simulation results show that the algorithm can effectively detect frequency hopping signals and blind parameter estimation in complex electromagnetic environment.3.Aiming at the problem of low automation in the recognition process of frequency hopping signal,combining with the existing image classification technology,the automatic classification and recognition of frequency hopping signal is realized by using frequency hopping signal classification algorithm based on time-frequency image gradient feature extraction.Firstly,UAV signals are converted in time-frequency domain.Pixel-by-pixel gradient feature extraction is carried out for samples of time-frequency waterfall map.Support vector machine is used as classifier.Kernel function parameters are optimized by genetic algorithm combined with cross-validation to realize the classification and recognition of frequency-hopping signals.This method does not need a priori condition of known target signal parameters.Blind classification is highly automated;when the total bandwidth and frequency hopping sequence period of the sample signal are known,an optimization algorithm for fixed window detection is proposed.The algorithm intercepts the single period of the target signal with a fixed window and extracts the features of the data in the window.In the detection module,the two-class detection is carried out by scanning the same window.The simulation results show that the method has higher recognition performance.The accuracy and classification effect are optimized.