Remote Sensing Image Aircraft Detection Based on Feature Fusion across Deep Learning Framework
Wanjun Wei, Jiuwen Zhang, Chengyu Xu · 2019
The detection of remote sensing image aircraft targets based on deep learning has practical and important significance in the fields of military reconnaissance and disaster rescue. As a typical representative of the two mainstream detection algorithms, YOLOv3 and Faster_R_CNN have good detection effects on remote sensing image aircraft targets. However, for low quality remote sensing images, the two detection algorithms also have the phenomenons of omission and false detection. To deal with this, this paper proposes a target detection algorithm (YF_R_CNN) for "Separate training, joint detection", which realizes the cross-platform detection feature fusion of YOLOv3 based on darknet framework and Faster_R_CNN based on tensorflow framework, effectively alleviating the problems of existing algorithms. The experimental results show that the detection accuracy of YF_R_CNN algorithm reaches 94.8%, which is 3.7% and 3.1% higher than YOLOv3 and Faster_R_CNN respectively. The detection accuracy is obviously improved, and the algorithm has better flexibility and robustness.