Automatic air-to-ground target information extraction method based on small sample training video

Abstract

The invention discloses an automatic air-to-ground target information extraction method based on a small sample training video. The method comprises the steps of image recognition, image preprocessing, image segmentation, target feature extraction, target classification, regular learning and error detection. When the recognition result is wrong, secondary training is carried out, and learning anderror detection are carried out again, so that after the unmanned aerial vehicle and other aircrafts acquire videos of people, vehicles, animals and the like on the ground, corresponding types are extracted and classified and analyzed to obtain corresponding types, movement directions, speeds and other information, and the reconnaissance application value of the aircrafts is improved.

Description

technical field [0001] The invention relates to an information acquisition method, in particular to an automatic extraction method of air-to-ground target information based on a small-sample training video. Background technique [0002] In recent years, with the development of low-altitude detection aircraft such as drones and sounding balloons, especially the rapid development of drones, the number of civilian drones and police drones has shown an explosive growth trend. It is estimated that the number of civilian UAVs will reach more than 3 million by 2020; UAVs are widely used in low-altitude reconnaissance and surveillance, which makes low-altitude air-to-ground video intelligence more and more important in low-altitude reconnaissance and surveillance. application. [0003] At present, most of the video acquisitions used by drones are 30x optical zoom and 1080P pods. Since most UAVs fly at a height of more than 150 meters, especially types of UAVs such as vertical take...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to provide a kind of automatic information extraction method based on neural network RCNN adopting OFSBL operator to train low-altitude air-to-ground video small and slow targets with small samples, aiming at solving low-altitude aircraft acquisition resolution is not high, vertical angle of view is large, Low-altitude air-to-ground video and small slow target intelligence under the unfavorable factors of high flying altitude, fast speed, small target pixel density, lack of air-to-ground sample library, few target features, and short duration of the target appearing in the video Extraction Method

Images