| Broad-spectrum multi-band detection is the basic method of high-precision detection systems.How to use image fusion technology to further improve the detection effect of the system in complex scenes is one of its bottlenecks.Due to the limitations of imaging sensors and other hardware devices and the real-time requirements of image information processing,the source images obtained often have low resolution,and the fusion results are not satisfactory.For low-resolution detection imaging,the existing research mostly focuses on two images,using the methods of "first super-resolution and then fusion" and "first fusion and then super-resolution" for processing.This paradigm of separation often leads to information loss or introduces noise,especially for the fusion of multiple images.Therefore,it is urgent to explore a new method of end-to-end multi-band image synchronization super-resolution and fusion.This article has done the following work based on deep learning:(1)A multi-band low-resolution image super-resolution and fusion method based on WGAN-GP is proposed.In order to still ensure the rapid and accurate extraction of the target during low-resolution imaging,firstly,the multi-band low-resolution source image is enlarged to the target size based on the bicubic interpolation method;Secondly,input the amplified result into the feature extraction(coding)network to extract the features separately and combine them in the high-level feature space;Then,the preliminary fusion image is obtained through the image reconstruction(decoding)network;finally,the high-resolution fusion image is obtained through the dynamic game of the generator and the discriminator.Experiments show that the proposed method can not only achieve multi-band image super-resolution and fusion simultaneously,but also the information,clarity and visual quality of the results are significantly higher than other representative methods.(2)A multi-band low-resolution image synchronization super-resolution and fusion method based on dense network and local brightness enhancement operator is proposed.Aiming at the problems of high cost in the process of multi-band image fusion and image super-resolution,high network computational complexity and slow processing speed,the dense network is used to construct a feature extraction module to extract the source image information of each band separately,and then a feature fusion module is constructed to combine the information of each band,and finally the image super-resolution is achieved through sub-pixel convolution,and the fusion result is obtained.In addition,the loss function is improved by using the local brightness boosting operator.Experimental results show that the proposed method can improve network efficiency and obtain high-resolution fusion results.(3)An unsupervised synchronous fusion method of multi-band images based on multi-discriminator generation adversarial network is proposed.Aiming at the problem of lack of label images when using deep learning to fuse multi-band images,first,design and construct a feedback dense network as a feature enhancement module to extract the features of each band source image and perform feature enhancement;Secondly,merge and connect the feature enhancement results and reconstruct the fused image through the designed feature fusion network module;Finally,the preliminary fusion results and the source images of each band are input into the discriminant network,and the generator is continuously optimized through the classification tasks of multiple discriminators,so that the output result of the generator retains the characteristics of multiple band images at the same time to achieve image fusion the goal of.The experimental results show that,compared with the current representative fusion methods,the proposed method has better clarity,more information,richer details,and is more in line with human visual characteristics.It is for exploring unsupervised learning to synchronize super-resolution and fusion method lays the foundation. |
PDF Full Text Request