Adversarial Attacks Against Object Detection in Remote Sensing Images

Rong Huang, Li Chen, Jun Zheng, Quanxin Zhang, Xiao Yu

Abstract: With the continuous development of artificial intelligence technology and the increasing richness of remote sensing data, deep convolutional neural networks(DNNs) have been widely used in the field of remote sensing images. Object detection in remote sensing images has achieved considerable progress due to DNNs. However, DNNs have shown their vulnerability to adversarial attacks. The object detection models in remote sensing images also have this vulnerability. The complexity of remote sensing object detection models makes it difficult to implement adversarial attacks. In this work, we propose an adversarial attack method against the remote sensing object detection model based on the \(L_{\infty }\)norm, which can make the detector blind–that is, the detector misses a large number of objects in the image. Because some remote sensing images are too large, we also designed a pre-processing method to segment and pre-process the huge images, which is combined with the attack method. Our proposed attack method can effectively perform adversarial attacks on remote sensing object detection models.