Searching And Tracking Technology Of "Low,Small And Slow" Target In The Air For The Safety Protection Of Important Places

With the development of remote sensing image processing technology,the search and tracking of "low,small and slow" targets in the air has an urgent demand for military and civilian use and a wide range of application prospects,which has become one of the key research issues in various countries.With the wide application of artificial intelligence,the target tracking technology for ordinary remote sensing video has been relatively mature.However,due to the problems of "low,small and slow" target,such as less pixels,easy to be interfered by similar targets around,and difficult to extract features,the current mainstream methods can not provide effective solutions for such targets.In this paper,aiming at the relevant difficulties,a complete set of "low small slow" target search and tracking software is constructed by using the methods of hybrid countermeasure network,time-space residual module,significance enhancement mechanism,context aware correlation filtering module,etc.,to carry out the research of "low small slow" target search and tracking in the air.This paper can be condensed into the following aspects:First of all,this paper proposes a method of sample augmentation of air "low small slow" targets based on hybrid countermeasure network.Aiming at the interference problem of "low small slow" targets with few samples,this method uses multi-level pyramid to augment the samples with multi-scale and multi rotation angles,and processes the obtained sample slices through a series of morphological methods for subsequent network training Training provides effective data support.Based on the prior information,this method can effectively identify the unusual "low,small and slow" interference targets in the air,and the effectiveness of the proposed hybrid countermeasure network is verified by experiments.Secondly,aiming at the structure of the network layer,this paper constructs an effective mechanism of significance enhancement.Aiming at the problem that "low,small and slow" target is difficult to search,this method designs space-time residualmodule and space attention,channel attention lightweight sub module,and optimizes its arrangement between convolution layers,so as to effectively improve the training and tracking effect of the network.Thirdly,in view of the occlusion in complex large field of view images,this paper uses context aware correlation filtering and reliability learning module to reduce the probability of tracking failure.Through the analysis of the target itself and background information,this method embeds two modules into the "interference stage" of the network,and optimizes the characteristics of the "low,small and slow" target to improve the tracking effect.After that,for the problem that "low,small and slow" targets usually occupy less pixels,this method uses an effective screening mechanism to eliminate negative samples with little training value,so as to effectively improve the efficiency of network search and tracking.Finally,on the basis of the above-mentioned multiple feature learning mechanisms,this paper designs a complete set of "low small slow" target search and tracking software,which realizes the accurate and fast identification of such targets in the complex large field of view environment.In this paper,with the support of self-made "low,small and slow" data set,qualitative and quantitative analysis is carried out,and multi-level and multi angle verification means are used to verify the effectiveness of the method proposed in this paper.