Research Article
Aircraft Detection in Aerial Remote Sensing Images Based on Contrast Self-supervised Learning
@INPROCEEDINGS{10.1007/978-3-031-04409-0_26, author={Yuanyuan Liu}, title={Aircraft Detection in Aerial Remote Sensing Images Based on Contrast Self-supervised Learning}, proceedings={Machine Learning and Intelligent Communications. 6th EAI International Conference, MLICOM 2021, Virtual Event, November 2021, Proceedings}, proceedings_a={MLICOM}, year={2022}, month={5}, keywords={Aerial remote sensing images Small sample data set ZL method Target detection}, doi={10.1007/978-3-031-04409-0_26} }- Yuanyuan Liu
Year: 2022
Aircraft Detection in Aerial Remote Sensing Images Based on Contrast Self-supervised Learning
MLICOM
Springer
DOI: 10.1007/978-3-031-04409-0_26
Abstract
UAV aerial remote sensing images, which have the advantages of high resolution, convenient acquisition, high amount of information, and simple pre-processing process, are widely used in classification and detection tasks. The difficulty in the study of aircraft target detection in remote sensing images is that it relies on a large amount of labeled data and the target is relatively small. Therefore, this paper firstly studies the existing comparative learning methods in the self-supervised field, and then proposes the ZL (a comparative learning method in the remote sensing field) method for the small sample remote sensing data set to realize the extraction of the representation of the aircraft target. The ZL method mainly modifies the data enhancement combination form in data augmentation pipeline of the training process and the fine-tuning process and modifies the activation function in the projection head. Finally, the ZL trained model is combined with Faster R-CNN to improve the accuracy of aircraft target detection in aerially remote sensing images.
