Research And Implementation Of Remote Sensing Images Super-resolution And Semantic Segmentation Method Based On Deep Neural Network

In recent years,with the rapid development of remote sensing technology,remote sensing images have become important Earth observation data and are widely used in fields such as environmental monitoring and urban planning.However,remote sensing images often have problems such as low resolution and susceptibility to noise interference.These problems not only reduce the visualization effect of remote sensing images,but also limit their analysis and application capabilities.Remote sensing image super-resolution reconstruction is one of the methods to solve the problems of low resolution and image degradation in remote sensing images.Among them,traditional remote sensing image super-resolution reconstruction methods have shortcomings such as low processing efficiency,difficulty in processing complex scenes,and difficulty in ensuring the quality of processing results.The super-resolution reconstruction method based on deep neural networks has high adaptability and generalization ability,which can automatically learn image feature representations from a large number of samples and improve the processing effect of remote sensing images.The existing methods for super-resolution reconstruction of remote sensing images based on convolutional neural networks usually use deep residual networks to extract features and upsampling methods to reconstruct highresolution images.However,the existing methods for super-resolution reconstruction of remote sensing images based on convolutional neural networks have high complexity,insufficient processing of image edge information,and are prone to blurring effects.Moreover,existing super-resolution reconstruction methods for remote sensing images often have issues with the separation of super-resolution processes from subsequent applications such as detection and segmentation tasks.In response to the above issues,the specific work of this article is as follows:(1)In response to the high complexity of existing image super-resolution models based on convolutional neural networks,unsatisfactory results in image edge information processing,and the tendency to produce blurring effects,this paper proposes a remote sensing image super-resolution reconstruction network based on lightweight attention mechanism.This network performs zero filling operations on the image before each convolution to solve the problem of image size reduction caused by gradual convolution.At the same time,lightweight Enhanced Channel Attention(ECA)is introduced into the network.Adding ECA can effectively improve the efficiency of model feature extraction,reduce model complexity,and enhance the model’s attention to high-frequency details and textures of images,extracting more critical image feature information.(2)This paper proposes a remote sensing image super-resolution reconstruction and semantic segmentation network based on parallel decoders to address the problem of task fragmentation in remote sensing image super-resolution reconstruction and semantic segmentation.The encoder part of the network uses a residual network based on lightweight attention mechanism for feature extraction,and the decoder decodes semantic segmentation and super-resolution reconstruction.The remote sensing image super-resolution reconstruction and semantic segmentation network based on parallel decoders parallelizes the decoders and introduces feature affinity loss to help super-resolution semantic segmentation learn high-resolution representations.(3)Based on the above research content,this article designs and implements a remote sensing image super-resolution reconstruction and semantic segmentation system.The system uses low resolution remote sensing images as input and utilizes the method proposed in this article to achieve super-resolution reconstruction and semantic segmentation functions.The operation is simple and interactive,and it has high application value.