针对图像盲超分辨率网络计算参数多、模型庞大的问题,对快速且节省内存的轻量级图像非盲超分辨率网络(fast and memory-efficient image super resulotion network,FMEN)进行改进,提出了一种轻量级的快速且节省内存的图像盲超分辨率网络...针对图像盲超分辨率网络计算参数多、模型庞大的问题,对快速且节省内存的轻量级图像非盲超分辨率网络(fast and memory-efficient image super resulotion network,FMEN)进行改进,提出了一种轻量级的快速且节省内存的图像盲超分辨率网络(fast and memory-efficient image blind super resulotion network,FMEBN)。首先,通过图像退化模块模拟部分真实世界退化空间,使用退化预测模块预测低分辨率(low resolution,LR)图像的退化参数;然后,为能有效利用退化先验信息指导并约束网络进行重建,使用动态卷积对原网络特征提取、重建模块、高频注意力块(high frequency attention block,HFAB)结构进行改进;最后,使用生成对抗网络(generative adversarial network,GAN)对FMEN训练策略与损失函数进行优化,减小真实数据与生成数据的差异,生成更加真实、清晰的纹理、轮廓。实验结果表明,在合成图像数据集和真实图像数据集RealWorld-38上,该算法有较好的重建精度与视觉效果,模型大小12 MB,可以满足图像盲超分辨率网络的轻量级需求。展开更多
Vegetation is crucial for wetland ecosystems.Human activities and climate changes are increasingly threatening wetland ecosystems.Combining satellite images and deep learning for classifying marsh vegetation communiti...Vegetation is crucial for wetland ecosystems.Human activities and climate changes are increasingly threatening wetland ecosystems.Combining satellite images and deep learning for classifying marsh vegetation communities has faced great challenges because of its coarse spatial resolution and limited spectral bands.This study aimed to propose a method to classify marsh vegetation using multi-resolution multispectral and hyperspectral images,combining super-resolution techniques and a novel self-constructing graph attention neural network(SGA-Net)algorithm.The SGA-Net algorithm includes a decoding layer(SCE-Net)to preciselyfine marsh vegetation classification in Honghe National Nature Reserve,Northeast China.The results indicated that the hyperspectral reconstruction images based on the super-resolution convolutional neural network(SRCNN)obtained higher accuracy with a peak signal-to-noise ratio(PSNR)of 28.87 and structural similarity(SSIM)of 0.76 in spatial quality and root mean squared error(RMSE)of 0.11 and R^(2) of 0.63 in spectral quality.The improvement of classification accuracy(MIoU)by enhanced super-resolution generative adversarial network(ESRGAN)(6.19%)was greater than that of SRCNN(4.33%)and super-resolution generative adversarial network(SRGAN)(3.64%).In most classification schemes,the SGA-Net outperformed DeepLabV3+and SegFormer algorithms for marsh vegetation and achieved the highest F1-score(78.47%).This study demonstrated that collaborative use of super-resolution reconstruction and deep learning is an effective approach for marsh vegetation mapping.展开更多
Visual cortical prostheses have the potential to restore partial vision. Still limited by the low-resolution visual percepts provided by visual cortical prostheses, implant wearers can currently only ＂see＂ pixelized...Visual cortical prostheses have the potential to restore partial vision. Still limited by the low-resolution visual percepts provided by visual cortical prostheses, implant wearers can currently only ＂see＂ pixelized images, and how to obtain the specific brain responses to different pixelized images in the primary visual cortex（the implant area） is still unknown. We conducted a functional magnetic resonance imaging experiment on normal human participants to investigate the brain activation patterns in response to 18 different pixelized images. There were 100 voxels in the brain activation pattern that were selected from the primary visual cortex, and voxel size was 4 mm × 4 mm × 4 mm. Multi-voxel pattern analysis was used to test if these 18 different brain activation patterns were specific. We chose a Linear Support Vector Machine（LSVM） as the classifier in this study. The results showed that the classification accuracies of different brain activation patterns were significantly above chance level, which suggests that the classifier can successfully distinguish the brain activation patterns. Our results suggest that the specific brain activation patterns to different pixelized images can be obtained in the primary visual cortex using a 4 mm × 4 mm × 4 mm voxel size and a 100-voxel pattern.展开更多
Existing methods of measurement MTF for discrete imaging system are analysed. A slit target is frequently used to measure the MTF for an imaging system. Usually there are four methods to measure the MTF for a discrete...Existing methods of measurement MTF for discrete imaging system are analysed. A slit target is frequently used to measure the MTF for an imaging system. Usually there are four methods to measure the MTF for a discrete imaging system by using a slit. These methods have something imperfect respectively. But for the discrete imaging systems of under sampling it is difficult to reproduce this type of target properly since frequencies above Nyquist are folded into those below Nyquist, resulting in aliasing effect. To tackle the aliasing problem, a super resolution technique is introduced into our measurement, which gives MTF values both above and below Nyquist more accurately.展开更多
Machine learning methods, one type of methods used in artificial intelligence, are now widely used to analyze two-dimensional (2D) images in various fields. In these analyses, estimating the boundary between two regio...Machine learning methods, one type of methods used in artificial intelligence, are now widely used to analyze two-dimensional (2D) images in various fields. In these analyses, estimating the boundary between two regions is basic but important. If the model contains stochastic factors such as random observation errors, determining the boundary is not easy. When the probability distributions are mis-specified, ordinal methods such as probit and logit maximum likelihood estimators (MLE) have large biases. The grouping estimator is a semiparametric estimator based on the grouping of data that does not require specific probability distributions. For 2D images, the grouping is simple. Monte Carlo experiments show that the grouping estimator clearly improves the probit MLE in many cases. The grouping estimator essentially makes the resolution density lower, and the present findings imply that methods using low-resolution image analyses might not be the proper ones in high-density image analyses. It is necessary to combine and compare the results of high- and low-resolution image analyses. The grouping estimator may provide theoretical justifications for such analysis.展开更多
文摘针对图像盲超分辨率网络计算参数多、模型庞大的问题,对快速且节省内存的轻量级图像非盲超分辨率网络(fast and memory-efficient image super resulotion network,FMEN)进行改进,提出了一种轻量级的快速且节省内存的图像盲超分辨率网络(fast and memory-efficient image blind super resulotion network,FMEBN)。首先,通过图像退化模块模拟部分真实世界退化空间,使用退化预测模块预测低分辨率(low resolution,LR)图像的退化参数;然后,为能有效利用退化先验信息指导并约束网络进行重建,使用动态卷积对原网络特征提取、重建模块、高频注意力块(high frequency attention block,HFAB)结构进行改进;最后,使用生成对抗网络(generative adversarial network,GAN)对FMEN训练策略与损失函数进行优化,减小真实数据与生成数据的差异,生成更加真实、清晰的纹理、轮廓。实验结果表明,在合成图像数据集和真实图像数据集RealWorld-38上,该算法有较好的重建精度与视觉效果,模型大小12 MB,可以满足图像盲超分辨率网络的轻量级需求。
基金supported by National Natural Science Foundation of China:[Grant Number 21976043,42122009]Guangxi Science&Technology Program:[Grant Number GuikeAD20159037]+1 种基金‘Ba Gui Scholars’program of the provincial government of Guangxi,and the Guilin University of Technology Foundation:[Grant Number GUTQDJJ2017096]Innovation Project of Guangxi Graduate Education:[Grant Number YCSW2022328].
文摘Vegetation is crucial for wetland ecosystems.Human activities and climate changes are increasingly threatening wetland ecosystems.Combining satellite images and deep learning for classifying marsh vegetation communities has faced great challenges because of its coarse spatial resolution and limited spectral bands.This study aimed to propose a method to classify marsh vegetation using multi-resolution multispectral and hyperspectral images,combining super-resolution techniques and a novel self-constructing graph attention neural network(SGA-Net)algorithm.The SGA-Net algorithm includes a decoding layer(SCE-Net)to preciselyfine marsh vegetation classification in Honghe National Nature Reserve,Northeast China.The results indicated that the hyperspectral reconstruction images based on the super-resolution convolutional neural network(SRCNN)obtained higher accuracy with a peak signal-to-noise ratio(PSNR)of 28.87 and structural similarity(SSIM)of 0.76 in spatial quality and root mean squared error(RMSE)of 0.11 and R^(2) of 0.63 in spectral quality.The improvement of classification accuracy(MIoU)by enhanced super-resolution generative adversarial network(ESRGAN)(6.19%)was greater than that of SRCNN(4.33%)and super-resolution generative adversarial network(SRGAN)(3.64%).In most classification schemes,the SGA-Net outperformed DeepLabV3+and SegFormer algorithms for marsh vegetation and achieved the highest F1-score(78.47%).This study demonstrated that collaborative use of super-resolution reconstruction and deep learning is an effective approach for marsh vegetation mapping.
基金supported by the National Natural Science Foundation of China,No.31070758,31271060the Natural Science Foundation of Chongqing in China,No.cstc2013jcyj A10085
文摘Visual cortical prostheses have the potential to restore partial vision. Still limited by the low-resolution visual percepts provided by visual cortical prostheses, implant wearers can currently only ＂see＂ pixelized images, and how to obtain the specific brain responses to different pixelized images in the primary visual cortex（the implant area） is still unknown. We conducted a functional magnetic resonance imaging experiment on normal human participants to investigate the brain activation patterns in response to 18 different pixelized images. There were 100 voxels in the brain activation pattern that were selected from the primary visual cortex, and voxel size was 4 mm × 4 mm × 4 mm. Multi-voxel pattern analysis was used to test if these 18 different brain activation patterns were specific. We chose a Linear Support Vector Machine（LSVM） as the classifier in this study. The results showed that the classification accuracies of different brain activation patterns were significantly above chance level, which suggests that the classifier can successfully distinguish the brain activation patterns. Our results suggest that the specific brain activation patterns to different pixelized images can be obtained in the primary visual cortex using a 4 mm × 4 mm × 4 mm voxel size and a 100-voxel pattern.
文摘Existing methods of measurement MTF for discrete imaging system are analysed. A slit target is frequently used to measure the MTF for an imaging system. Usually there are four methods to measure the MTF for a discrete imaging system by using a slit. These methods have something imperfect respectively. But for the discrete imaging systems of under sampling it is difficult to reproduce this type of target properly since frequencies above Nyquist are folded into those below Nyquist, resulting in aliasing effect. To tackle the aliasing problem, a super resolution technique is introduced into our measurement, which gives MTF values both above and below Nyquist more accurately.
文摘Machine learning methods, one type of methods used in artificial intelligence, are now widely used to analyze two-dimensional (2D) images in various fields. In these analyses, estimating the boundary between two regions is basic but important. If the model contains stochastic factors such as random observation errors, determining the boundary is not easy. When the probability distributions are mis-specified, ordinal methods such as probit and logit maximum likelihood estimators (MLE) have large biases. The grouping estimator is a semiparametric estimator based on the grouping of data that does not require specific probability distributions. For 2D images, the grouping is simple. Monte Carlo experiments show that the grouping estimator clearly improves the probit MLE in many cases. The grouping estimator essentially makes the resolution density lower, and the present findings imply that methods using low-resolution image analyses might not be the proper ones in high-density image analyses. It is necessary to combine and compare the results of high- and low-resolution image analyses. The grouping estimator may provide theoretical justifications for such analysis.
