Abstract
In order to improve the accuracy of target tracking from aerial remote sensing image and reduce the interference of background, a target tracking method based on reliability learning multi-domain convolution neural network (RLMD) is proposed. Firstly, in the stage of network architecture designing, the optimized residual framework is employed to enhance the saliency of target features. Secondly, in the stage of online tracking, an adaptive reliability learning module is introduced into the tracking network when the predicted score is in the interference stage. The tracking performance is improved by suppressing the response of background area in the target frame. Finally, the qualitative and quantitative experiments are conducted on the UAV123 and self-made datasets. The results show that RLMD algorithm can achieve high accuracy and high timeliness tracking performance for targets from aerial remote sensing platforms.
| Translated title of the contribution | Remote Sensing Video Target Tracking Based on Reliability Learning Multi-Domain Convolutional Neural Network |
|---|---|
| Original language | Chinese (Traditional) |
| Pages (from-to) | 98-102 |
| Number of pages | 5 |
| Journal | Beijing Ligong Daxue Xuebao/Transaction of Beijing Institute of Technology |
| Volume | 39 |
| Publication status | Published - Oct 2019 |