Research On FOD Detection Methods In Ground Clutter Of Air Landing Strips

Foreign object debris(FOD)is defined as such items that appear on an airport runway and may cause damage to aircraft.At present,manual inspection combined with optical sensors are played as the main method against FOD,which depend on weather and light conditions.In contrast,radars have received widespread attention due to working all-day and all-weather.This paper focuses on the radar-based FOD detection under low signal to clutter ratio conditions of airport runway environment,and systematically studies the ground clutter characteristics as well as target detection methods under clutter background.Chapter One first explains the research background and significance,analyzes the hazards of FOD to aviation safety and the urgent requirement for FOD detection technology.The performance of the current FOD detection systems are evaluated,and the key technology and development history and current situation of FOD radar detection are also investigated.At the end,the research ideas and approaches are given and explained.The second chapter studies statistical modeling method of ground clutter based on the scattering characteristic analysis of airport runway.First,the empirical model and fitting method of the runway backscattering coefficients are introduced.Based on the measured data,the optimal combinations of model orders that conform the airport runway fact are analyzed.In order to describe the probability density of complex clutter from different scattering surfaces,a compound Weibull distribution model is proposed.The convergence of analytical solutions is difficult to obtain by traditional parameter estimation methods,even numerical solutions are hardly guaranteed in the definitional domain.Aiming at the problems above,a method based on moment estimation and Newton-Raphson iteration is proposed,which shows higher efficiency and fewer iterations.At the end of this chapter,the half-space experimental platform in a darkroom is introduced aiming at data measurement.The experimental results based on darkroom measurement data reveal that the clutter power of a single environment obeys the traditional Weibull distribution with two parameters.The compound Weibull model can describe the probability density of complex clutter from different scattering surfaces,in addition,the effectiveness of the proposed parameter estimation method is verified.Chapter Three investigates the FOD detection method based on clutter map-constant false alarm rate(CM-CFAR).As the basis,the performance of CM-CFAR surface technology and are analyzed and compared with point technology.Aiming at the detection difficulty of dim targets at clutter edge,a cell-averaging CM-CFAR detector based on variability index(VI)is proposed.In homogeneous background,the threshold is calculated by the student-t-distributed test statistic;under the discontinuous clutter conditions,the threshold is modified according to current VI,in order to address the performance decrease caused by extended clutter edges.A joint multi-frame CM-CFAR detector with double thresholds is presented: double thresholds are utilized according to non-zero VI in range or azimuth,which indicate clutter edges,and the multi-frame judgment results are merged to achieve the presence or absence of dim FOD targets especially at clutter edges.Finally,the experiment results by measured data support such conclusions: detectability improvement are obtained by the above two CM-CFAR to dim items,compared with traditional CM-CFAR.The fourth chapter studies the knowledge aided space-time adaptive processing(STAP)technology for moving FOD indication,which is supported by the knowledge of runway environment.for the problem of moving FOD detection.Firstly,a side-looking observation model is established,the scene is divided by rectangular virtual scattering units,thus a clutter covariance matrix estimation method aided by the knowledge of runway environment is proposed,which improves the STAP performance degradation caused by insufficient independent and identically distributed samples in inhomogeneous clutter environment.Then the KA-STAP algorithm is introduced and evaluated by such performance indexes: the loss function,the improvement factor,as we as the minimum detectable Doppler.The virtual element redundancy of the multiple-input-multiple-output system may cause the space-time resolution loss,the minimum redundancy array based STAP is proposed,which reduces the system processing dimension and improves the output signal to clutter-noise ratio of system.Besides,another KA-STAP method based on an expanded difference array is proposed,and the performance are researched with three non-uniform linear arrays as examples.The simulation shows that the effectiveness and cost-effectiveness advantages of the presented KA-STAP methods in clutter suppression and moving target indication.On the basis of deep neural networks,Chapter Five focuses on runway environment perception and FOD detection.First,the clutter sensing method based on convolutional neural networks(CNN)is studied,and the influence of hyperparameters on network is analyzed.A data reconstruction method based on sliding window is constructed,for input sequence enhancement,the improvement of classification accuracy,network convergence efficiency is verified by experiment,moreover,this method can reduce the influence of initial iteration.Clutter statistic modelling is limited by insufficient samples,thus the conditional variational auto-encoder(CVAE)is used to develop a clutter statistical model reconstruction\generation network.Correspondingly,the employment of grazing-angle labels is improved to obtain more reliable generated model.Transfer learning play as the theory basis for FOD detection.In detail,a typical target is used to train the source CNN,thus some FOD targets can be detected through network transfer and necessary finetune(to reduce the differences between source and target domain).Besides,the multi-scale kernels are employed in CNN and analyzed.The proposed detection method are verified by measured data,and the network structure shows better performance to indicate FOD targets in clutter background.Chapter Six summarizes the research work and points out the problems to be further studied.