Moving Point Target Detecting Based On High Frame-rate Image Sequence In Very Low SNR

Visual detection of moving point target is an increasingly important tast in military application and space target detection.Detection approaches have been changing from ground-based to satellite-based.Target size is becoming smaller and smaller while the moving speed is becoming higher and higher.The high moving speed and small size of specific targets such as space debris and hypersonic vehicle,are of great changlleges for targt detection.With the development of remote sensing technology,a new high-precision,real-time systm is being built to estabilish an information fusion network of sea,ground,sky and space.Remote sensing methods have been changing form static to dynamic,ground-based to satellite-based and local area to the global area.Multi-sensor,multi-platform and multi-angle remote sensing methods are designed,and high space-resolution,hyperspectral,high time-resolution and high radiative-resolution payloads are in developing.These new information acquisition approaches must lead to new data processing methods.In this thesis,we are focusing on the moving point target detection using high frame-rate image sequences.After the research of existing moving target detection methods,a novel high frame-rate based moving point targe detection framework is built.The main contributions of this thesis are as follows.A high frame-rate based moving point target detection framework is built to detect fast moving point target in very low SNR,in which the main domain of interest changes form space to time.Unlike infrared image,this thesis foucs on visual image,in which the SNR is too low for the taget to be detected in space domain.With the development of high-speed imaging techniques,it became possible to exploit statistical methods more efficiently for point target detection.The idea relies on the fact that with high-speed imaging there will be a substantial number of image samples in a very short time,with each pixel's observed intensity in temporal domain forming a time series,and the higher number of samples enables more accurate statistical modelling of the signal.The key detection problem is formulated to detect a time-domain transient signal of unknown scale and arrival time in noisy background.The mathematical model of the moving point target and the target detector model are proposed in this thesis,which follows by a time domain background modeling method.The detection ability is calculated in time domain.Theoretical analysis and simulation result show that the detection ability is proportional to the frame-rate.A higher-order statsitcs method is proposed for moving point target detection in stationary and noisy background under the high frame-rate based framework.A novel target detector based on bispectral is developed,which uses a multi-variable test method of mean and variance in bispectral domain.The time domain background modeling is used to calculate the key parameters in the detector.The detection ability under different frame-rates and SNRs is analyzed.The method is evaluated using both simulated and real-world data and we provide a comparison to other widely used point target detection approaches.The experimental results demonstrate that our algorithm can efficiently detect extremely low SNR targets that are virtually invisible to humans based on time-domain analysis of image sequences.A time domain correlation analysis method is proposed for transient signal detection.The wavelet transformation is used to suppress the background noise,which follows by a kernel-based target detector.The kernel-based method maps the time series from the original space to a high dimensional feature space,in which the target is amplified while the noise is suppressed.The proposed detector can serve as the target detection under the high frame-rate based framework,which is effective in target detection in gradually changing background.The experimental results demonstrate that the kernel-based target detector can efficiently detect extremely low SNR targets compared with other widely used methods.