Aircraft image recognition in airport flight area based on deep transfer learning

Abstract

Aircraft recognition is of great significance to the construction of intelligent airport surface video monitoring system. At present, there are few studies on aircraft identification in airport flight area. In order to realize the accurate monitoring of airport surface aircraft, the method of transfer learning is proposed to identify aircraft. By collecting the images of Sichuan Airlines aircraft in Chengdu Shuangliu international airport, a total of 539 data sets composed of five types of AirBus aircraft images are constructed. 80% images are randomly selected as the training set and 20% images as the test set. The method of constructing deep transfer learning network based on AlexNet, ResNet and EfficientNet is used to modify partial convolution layer and full connection layer for the obtained dataset. The accuracy rates of the three deep learning network models on the test set are 88.2 %, 98.1 % and 94.4 % respectively. The experimental results show that using deep transfer learning method to train small or medium sample aircraft image dataset can achieve high recognition accuracy.