Unmanned aerial vehicle front-end recognition system fused with low-power-consumption chip deep learning algorithm

Abstract

The invention provides an unmanned aerial vehicle front-end recognition system fused with a low-power-consumption chip deep learning algorithm, which comprises an adjusting device, an unmanned aerial vehicle, an acquisition device and a processor; the adjusting device adjusts the posture of the unmanned aerial vehicle; the acquisition device is used for acquiring data of an advancing path of the unmanned aerial vehicle; the acquisition devices are respectively arranged on the unmanned aerial vehicle; the acquisition device is arranged on the unmanned aerial vehicle and performs data acquisition on the operation front end of the unmanned aerial vehicle; the acquisition device comprises an acquisition mechanism and an acquisition algorithm; the acquisition mechanism is used for acquiring data in the operation process of the unmanned aerial vehicle; the acquisition algorithm is used for processing data acquired by the acquisition mechanism; the acquisition mechanism comprises an angle deflection component and an acquisition probe, and the angle deflection component is used for adjusting the detection position of the acquisition probe. According to the invention, the detection probe arranged at the front end of the unmanned aerial vehicle collects the image or video data, and carries out analysis according to the image or video data, thereby achieving the precise recognition of an object or an area.

Description

technical field [0001] The invention relates to the technical field of unmanned aerial vehicle equipment, in particular to an unmanned aerial vehicle front-end recognition system integrated with a low-power chip deep learning algorithm. Background technique [0002] The existing UAV detection and recognition methods use optical imaging sensors to automatically survey the sky to obtain image sequences of the area to be detected, and use the target motion characteristics between sequence images and the difference between the target and the background in a single image to detect UAVs. Wait for the low-altitude aircraft. [0003] Such as CN106707296B prior art discloses a kind of UAV automatic detection and identification method based on dual-aperture photoelectric imaging system, this method is susceptible to environmental interference, it is difficult to distinguish false alarms caused by UAV target and background interference. And after the target is detected, the target can...

AI Technical Summary

Problems solved by technology

[0003] Such as CN106707296B prior art discloses a kind of UAV automatic detection and identification method based on dual-aperture photoelectric imaging system, this method is susceptible to environmental interference, it is difficult to distinguish false alarms caused by UAV target and background interference

And after the target is detected, the target cannot be further identified based on the existing information

In addition, the existing technology is based on the implementation of radar-based monitoring of UAV targets, but there is still the problem of not being able to identify the target type. At the same time, the cost of radar equipment is high and it is susceptible to interference from factors such as weather and environment.

Existing technologies such as KR101844364B1, EP2416446B1, and US08747236B1 have been found through a large number of searches. Regarding the difference in micro-Doppler features of drones, more features that can be used to distinguish different types of drones can be extracted to improve the classification accuracy of drones, but the disadvantage is that it cannot detect and identify multiple targets. At the same time, the method of UAV detection is realized by using the UAV's velocity rhythm map (CVD), which uses the UAV's velocity rhythm map feature to make up for the lack of micro-Doppler features, and realizes the correct classification and identification of UAVs. The main disadvantage is that it cannot perform multi-target recognition and detection. At the same time, using the convolutional neural network as a classifier has high computational complexity and large time overhead, and cannot achieve real-time processing.

[0004] In order to solve the common problems in the field such as low recognition efficiency, non-adjustable recognition angle, inability to predict recognition features, inability to perform deep learning and automatic recognition, etc., the present invention is made

Images