A method to reduce false alarm probability in deep network detection

Abstract

The invention belongs to video SAR moving target detection and tracking technology, in particular to a method for reducing false alarm probability in deep network detection. The present invention proposes a method for reducing false alarms in deep network detection based on the combination of inter-frame information in traditional detection. First, a data set is created based on VideoSAR imaging images, and the speed range is calculated and the speed range is used as a label to form a data set for moving targets. Shadows are labeled and applied to deep networks for detection. The pre-trained model detects the subsequent frames in the video to obtain preliminary detection results, and then uses the detected speed range to obtain the prediction information of the preceding and following frames, and eliminates false alarm targets. Compared with the traditional background construction, the present invention does not lose the detection effect of the high-speed target, and at the same time, the false alarm probability is effectively suppressed on the basis of almost guaranteeing the detection probability.

Description

technical field [0001] The invention belongs to video SAR moving target detection and tracking technology, in particular to a method for reducing false alarm probability in deep network detection. Background technique [0002] Moving target detection and analysis has always been the key and hot spot in the research field of Synthetic Aperture Radar (SAR). The American Sandia laboratory proposed the VideoSAR imaging mode in 2003. Through the high frame rate and high resolution imaging of the scene, the dynamic observation of the ground can be realized, and the relevant information of the target area can be grasped in real time. [0003] At present, domestic research on VideoSAR moving target detection is still in its infancy, and obtaining moving targets from high frame rate video images and tracking them is the key technology. The motion of the target causes the image to be defocused, but the Doppler frequency shift causes it to leave a shadow in the actual position, so the...

AI Technical Summary

Problems solved by technology

[0006] Based on the CattePM model, the VideoSAR image is denoised, the coherent speckle noise in the VideoSAR image is removed, the anisotropic CattePM is used for denoising, the median method is used to construct the background model to achieve background separation, and the goal is achieved by the three-frame difference method The purpose of detection, but when the image is denoised and the background model is built, the shadow of the high-speed target is not obvious and it is easy to be flooded, so it cannot be detected

[0007] Based on the CNN network, the shadow generated by the target of the VideoSAR image is trained and recognized, because the shadow features generated by the target are less, and because there is a lot of noise in the VideoSAR image, and because there are a large number of static non-targets with low reflection coefficients in the detection area. False shadows , resulting in a large number of false alarm targets that are difficult to eliminate based on deep network detection

Images