Aircraft recognition using convolutional neural network
N. Wan Zulkipli, Joon Huang Chuah, Marcus Lim, G. M. T. Chai · IET conference proceedings. · 2023
Satellite image object-recognition offers detailed location representation where ground-based sampling is limited or inaccessible, such as high sea or in the mountains. However, an accurate and effective method is required to ensure the precision of the information generated due to small, ambiguous objects against the background. Having an automatic object detection system would greatly speed up satellite image recognition and find extensive applications in defense, transportation, agriculture, and emergency relief. With the advancement of artificial neural networks, the technology has seen applications in various fields from image processing to speech synthesis. We aimed to develop a neural network-based commercial aircraft detection system on satellite images at major airports. We proposed a R-CNN-based aircraft recognition system where CNN is utilized to extract features of aircraft before passing the feature maps to the binary classifier. We trained the model with 2100 images of satellite images containing aircraft which are segmented to produce positive and negative instances and validated on 900 images. The system is equipped to detect and localize the position of the aircraft in an image while registering excellent accuracy. However, as the number of aircraft in a frame increases, slight deterioration in the model performance is observed.