Improving SSD for detecting small target in Remote Sensing Image

摘要

This paper proposes an improved SSD algorithm to identify small targets in remote sensing images. After introducing the performance of the traditional neural network in target recognition, this article focuses on explaining the SSD algorithm, a new improved algorithm then is proposed to solve the weakness of the SSD algorithm. The proposed algorithm combines the top-to-down structure of the FPN with the SSD algorithm. Experimental results verify that the proposed method performs better than SSD in remote sensing images without increasing the amount of calculation.