Research On Small Target Detection Technology Based On Features

In the 21st century,no matter in the military or civil fields,various countries have gradually begun to pay attention to the protection of marine rights and interests,with new requirements for the accuracy and range of sea surface target detection.The detection ability of modern radar has many irreplaceable applications in the fields of territorial defense,ocean investigation,law enforcement in territorial waters,search and rescue in high seas,etc.At present,scholars at home and abroad have done a lot of researches on using electromagnetic wave to detect small floating targets on the ocean surface.These studies reveal that sea surface target detection still faces many challenges:firstly,sea clutter is the background of sea target,and the clutter intensity is often higher than the target signal intensity.Secondly,in order to obtain high signal to clutter ratio,radar will increase the bandwidth to improve the range resolution when detecting small sea surface targets,but sea clutter will show multi-dimensional non Gaussian characteristics under high resolution.Thirdly,the common floating small targets on the sea have similar motion patterns to the sea waves,which makes it difficult for radar to distinguish clutter and targets in the Doppler domain by phase reference processing.In this paper,based on the research status of small floating target detection on the sea,the technical routes and related algorithms are introduced.In the second chapter,the amplitude distribution models and nonlinear characteristics of sea clutter are introduced.Several kinds of amplitude distribution models of sea clutter which are most commonly used in engineering are introduced.The real sea clutter data sets are used to verify the characterization ability of various models for real clutter signals.Then,the radar target detection problem is treated as a binary classification problem,and three features used to distinguish clutter and target are selected to form a joint feature vector.In order to combine these features and distinguish target and clutter in feature space,a SVM classifier with adjustable false alarm rate is designed.Finally,referring to the structure of tracking before detection technology,the detector based on multi-dimensional features is introduced into the preprocessing part of tracking before detection algorithm.At the same time,considering that the radar target is generally unknown,we can regard the detection of small targets on the sea as an anomaly detection problem,and use the basic principle of principal component analysis algorithm to build an anomaly detector.In order to meet the requirements of radar system for false alarm rate,the error size is sorted,and the samples whose error value is greater than a certain threshold are judged as targets.Compared with the commonly used tracking before detection algorithm based on dynamic programming,the detector can improve the detection performance and still extract tracks with high reliability when the signal to clutter ratio decreases.