Research on Object Detection in Near-Infrared Remote Sensing Images Based on Feature Transfer
Authors:
- 7 July 2024
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 7752-7755
Abstract
Although visible light images are relatively easy to obtain, they are easily affected by weather and other environmental factors. Because of its high recognition capability of temperature changes, infrared remote sensing is more useful for target detection. However, the number of existing infrared remote-sensing image datasets is small, and the lack of training data will lead to a decline in accuracy. To solve this problem, this work proposes a detection method of ground objects based on feature transfer in infrared remote sensing images. A residual structure is used to solve the problem of network degradation, and a gradient inversion layer module is used to reduce the difference between domains, so as to realize feature transfer of feature maps in different levels and scales and fully consider the feature differences between global and local information. The experimental results indicate that the proposed network presents good performances on a famous challenging dataset.Keywords
This publication has 8 references indexed in Scilit:
- Transfer Learning in Deep Reinforcement Learning: A SurveyIEEE Transactions on Pattern Analysis and Machine Intelligence, 2023
- LIL: Lightweight Incremental Learning Approach Through Feature Transfer for Remote Sensing Image Scene ClassificationIEEE Transactions on Geoscience and Remote Sensing, 2021
- A Comprehensive Survey on Transfer LearningProceedings of the IEEE, 2021
- Deep gated attention networks for large-scale street-level scene segmentationPattern Recognition, 2019
- Domain Adaptive Faster R-CNN for Object Detection in the Wild2018
- Domain-Adversarial Training of Neural Networks2017
- Deep Residual Learning for Image Recognition2016
- Fast R-CNN2015