Dim And Small Air Target Detection In Complex Background

Small target detection is one of the most valuable research topics in the field of image processing.It is widely used in the detection and tracking of low-altitude airspace early warning systems and guided weapons such as missiles in long-range large field of view.The small target has a wide range of scale changes,less available information,and complex and variable scenes.How to effectively detect small targets appearing in the scene is an urgent technical solution for this research topic.The main research contents of the thesis are as follows:Aiming at the problem that traditional multi-stage filters can not adapt to multi-scale small target detection,a multi-scale cascade filtering fusion algorithm is proposed in this paper.Firstly,the background suppression of the original image is performed by using cascade filters of multiple scales.The residual image is obtained,and then multi-scale small target detection is realized by merging the residual images at each level.At the same time,combined with the image pyramid detection method to reduce the calculation amount of the algorithm.In view of the fact that the traditional side suppression algorithm can not protect the target edge very well,and the sliding window size is too large,which causes time-consuming problems.Two improvement measures are present: setting the isolation window to protect the target edge;redefining the 3*3 region to represent target region to reduce the calculation.Aiming at the problem of large number of redundant calculations in the background region based on local feature algorithm,an improved local contrast algorithm is put forward,which firstly determines the background point of the pixel to be measured,and then only performs grayscale enhancement on the non-background point;and the improved local contrast algorithm in the salient region is used to improve the performance of the algorithm.The experimental results show that the detection performance of the weak target detection algorithm based on local features is better than the filtering algorithm based on background prediction.The traditional target detection binary classification algorithm designs the feature artificially depending on prior knowledge,which is short of good adaptability.For this reason,the deep learning algorithm model is used to remedy these defects of the traditional binary classification algorithm mining the characteristics of the target and learning.In this paper,the lightweight network MobileNet V2 is used as the basic network to extract features,and combined with the SSD(Single Shot MultiBox Detector)detection framework,which solves the problem of small target detection,and realizes the adaptive target detection.