SIFT-Enhanced CNN Based Objects Recognition for Satellite Image
Kuang-Zhe Liu, Pei‐Jun Lee, Guocheng Xu, Bo-Hao Chang · 2020
Satellite images usually contain multiple in randomly located objects whose size is minimal since the height of the satellite observation. Because SIFT feature points have the invariant ability the same as CNN feature, to increase the accuracy of satellite images in object recognition, this paper uses the Scale-Invariant feature transform (SIFT) algorithm to locate the objects. Image cropping is applied to get the object region from image features. These images are put into CNN object recognition system. As experiment results, the proposed algorithm can recognize specific objects such as aircraft and ships with accuracy is about 70%.