An Airspace Security Situation Assessment Method Based on Class Activation Mapping Technology

Abstract

The invention discloses an airspace security situation assessment method based on category activation mapping technology, which belongs to the technical field of aircraft. Firstly, a data set is constructed for the historical airborne pictures collected by a drone, and the manual label processing is performed, and the labeled data set is divided into a training set, a verification set and a test set. Then construct the residual network structure as the main model of the perception network, and input the training set for training, and use the verification set for verification. Input the test set into the trained perception network main model, and end when the accuracy rate reaches more than 90%. For the actual flying drone, use the trained perception network main body model to judge and evaluate the current situation of the drone in real time. According to the real-time evaluation and visualization results of the image at each moment, the drone is guided away from the dangerous area and towards the safe area. The present invention is more intuitive, and adopts a simpler and simpler way to stipulate that the UAV is far away from the dangerous area.

Description

technical field [0001] The invention belongs to the technical field of aircraft, and relates to an airspace security situation assessment method based on category activation mapping technology. Background technique [0002] At present, the technologies related to the autonomous navigation of UAVs are divided into three parts: perception, decision-making and coordination according to their functions. In order to facilitate the subsequent packaging and deployment of the model, some teams are more accustomed to adopting an end-to-end approach, that is, using a neural network model to complete the above three parts. This end-to-end model generally takes the raw data collected by the drone as input, and then outputs a specific strategy or action. Although it can simplify the complexity of the operation, it also brings the problem of model interpretability. Using a neural network model to complete multiple tasks will make the model approach an unexplainable "black box". At the ...

AI Technical Summary

Problems solved by technology

This method can deal with uncertain information and can perform real-time evaluation, but this method pays too much attention to the transmission and accumulation of information, which makes the UAV unable to make better judgments on some emergencies

Images