Research On Multi-target Recognition Algorithm Of Remote Sensing Image Based On Super-resolution Reconstruction

With the rapid development of aerospace technology,remote sensing image processing has also become a research hotspot in the field of computer vision.The various target information contained in remote sensing images is of great significance for surface observation,and also greatly promotes the development of ecological monitoring,disaster control,and military and national defense.At this stage,most of the algorithms for remote sensing image target detection aimed at high-resolution remote sensing image datasets,and the recognition effect of these algorithms is often not ideal for low resolution remote sensing images with fog.For this problem,this paper proposes a multi-target recognition algorithm based on super-resolution reconstruction in remote sensing images,as described below.First,for low-resolution foggy remote sensing images,we propose an improved GAN super-resolution reconstruction algorithm for super-resolution reconstruction.We introduce a dense residual block into the network structure,and deletes the batch normalization layer in the residual block to make the generated remote sensing image texture more natural;secondly,the discriminator of the generative adversarial network is improved,using the relative The average discriminator allows the discriminator to discriminate the relative realism of the remote sensing images generated by the generator.The experimental results show that the algorithm proposed in this paper can reconstruct remote sensing images with rich texture details and edge information.Secondly,we implement a remote sensing image object detection algorithm based on the rotation equivariant feature for the reconstructed remote sensing image.This algorithm adds a rotation equivariant network to the CNN network to extract the rotation equivariant features,and proposes a rotation invariant Ro I Align,which can adaptively extract the rotation invariant features from the rotation equivariant features according to the direction of the Ro I.The experimental results show that the algorithm proposed in this paper can significantly improve the recognition of small targets with variable directions in remote sensing images.The multi-target recognition algorithm of remote sensing images based on superresolution reconstruction proposed in this paper improves the target recognition rate of remote sensing images in low-resolution foggy scenes,which proves the effectiveness of this method.