Research On Vision-based Recognition And Positioning Of Carrier-based Aircraft Model

As a simple,effective and intuitive form of hardware-in-the-loop simulation,hardware-in-the-loop simulation platform of the aircraft carrier has been used for the command and dispatch of carrier-based aircraft at home and abroad for a long time.It can provide significant guidance for the design of aircraft carriers.Using the carrier-based aircraft models to replace the real objects in the hardware-in-the-loop simulation platform,the efficiency and accuracy of the recognition and positioning are related to the operation of the whole simulation platform.Therefore,how to achieve a high accuracy and strong real-time carrier-based aircraft model recognition and positioning system is an urgent problem to be solved.Firstly,according to the characteristics of the hardware-in-the-loop simulation platform of the aircraft carrier,this thesis puts forward an overall scheme of recognition and positioning,splitting the total task into three parts to complete: target detection,number recognition,and multi-camera result fusion,and then analyzes and selects the hardware of the image acquisition module,and completes the hardware layout and the software design of the system;secondly,after analyzing the mainstream target detection algorithms,YOLOv3 is selected as the basic target detection algorithm,and then the design of the target detection algorithm of the carrier-based aircraft model is completed,and next based on the result of the target detection of the carrier-based aircraft model,the recognition of its number is completed;finally,using Visual Studio 2019 and Open CV3.4.0 and other tools,combined with coordinate transformation and camera calibration and other technologies,a vision-based recognition and positioning system of carrier-based aircraft model is realized.The main research contents and innovations of this thesis are as follows:(1)In view of the subject’s detection requirements for the direction angle prediction of the carrier-based aircraft model,adding the direction angle prediction and the multitask loss function with angle loss based on YOLOv3 to realize the above requirements;taking the network redundancy problem of the original network when facing the scene of the subject into account,the network structure is simplified and it speeds up the detection;in view of the tiny differences between the same or similar types of carrier-based aircraft models,adding a new detection scale to improve the fine-grained detection;in order to further improve the accuracy and the speed of convergence of target position prediction,the DIo U index is used to calculate the loss of the bounding box instead of the original algorithm’s Io U index;aiming at the problem of the small number of samples in self-built data set,transfer learning is applied to pre-train the model,which improves the generalization ability of the model.Experimental results show that the improved YOLOv3 algorithm can improve the m AP of carrier-based aircraft model detection by 9.5% and the detection speed by 5%.(2)In order to solve the problem of poor recognition accuracy of a single classifier in the identification of the number of carrier-based aircraft model,combining the characteristic of structural features and statistical features,a multi-level classifier for rotating numbers was designed.The experimental results show that the recognition rate and reliability of the number recognition of carrier-based aircraft model can be improved by 14.57% and 8.12% respectively.(3)In view of the possible rejection and misrecognition of the number recognition of carrier-based aircraft model,post-processing of the results is used to improve the reliability of recognition as much as possible.