Deep Super-resolution Network Research Articles

A novel method combining deep learning with the Kennard-Stone algorithm for training dataset selection for image-based rice seed variety identification.

Different varieties of rice vary in planting time, stress resistance, and other characteristics. With advances in rice-breeding technology, the number of rice varieties has increased significantly, making variety identification crucial for both trading and planting. This study collected RGB images of 20 hybrid rice seed varieties. An enhanced deep super-resolution network (EDSR) was employed to enhance image resolution, and a variety classification model utilizing the high-resolution dataset demonstrated superior performance to that of the model using the low-resolution dataset. A novel training sample selection methodology was introduced integrating deep learning with the Kennard-Stone (KS) algorithm. Convolutional neural networks (CNN) and autoencoders served as supervised and unsupervised feature extractors, respectively. The extracted feature vectors were subsequently processed by the KS algorithm to select training samples. The proposed methodologies exhibited superior performance over the random selection approach in rice variety classification, with an approximately 10.08% improvement in overall classification accuracy. Furthermore, the impact of noise on the proposed methodology was investigated by introducing noise to the images, and the proposed methodologies maintained superior performance relative to the random selection approach on the noisy image dataset. The experimental results indicate that both supervised and unsupervised learning models performed effectively as feature extractors, and the deep learning framework significantly influenced the selection of training set samples. This study presents a novel approach for training sample selection in classification tasks and suggests the potential for extending the proposed method to image datasets and other types of datasets. Further exploration of this potential is warranted. © 2024 Society of Chemical Industry.

CGIHE-VDSR: Color global image histogram equalization with very deep super resolution networks for color image super resolution

Medical and satellite image analysis require incredibly high resolution. Super-resolution combines several low-resolution images of the same scene to generate a high-resolution image. The Super resolution employing deep learning techniques still has an illumination issue. This paper proposes a novel CGIHE-VDSR algorithm that integrates the Very Deep Super Resolution (VDSR) Network with Color Global Image Histogram Equalization (CGIHE) to improve image resolution. In the proposed method, the low-resolution image is first histogram equalized using the CGIHE algorithm. Then, the VDSR network is applied to the histogram equalized image for super-resolution. The comparison of real-time data with the benchmark images is done using the proposed algorithm in the MATLAB platform. The PSNR and SSIM metrics demonstrate that the super resolution image obtained using the proposed method is significantly better than the existing methods.

Scene recovery: Combining visual enhancement and resolution improvement

Visibility enhancement of outdoor images under complex imaging conditions has been a crucial task for computer vision and received growing attention. However, existing image enhancement methods could result in typical block-like artifacts or color distortion. The undesirable impurities might also be significantly magnified after the enhancement task, further reducing the image quality. For enhancing and super-resolving complex real-world degradation, we propose a simultaneous visual enhancement and resolution improvement (VERI) variational scene recovery model for jointly enhancing image visibility and improving the resolution of the degraded image. Particularly, we estimate the scattering light map for degradation images to achieve clean scene radiance and simultaneously seek a high-quality image through a deep super-resolution network. The semi-proximal alternating direction method of multipliers (sPADMM) algorithm is employed for efficiently solving the minimization problems in the proposed model. Extensive experiments illustrate the effectiveness and robustness of the proposed method in dealing with various scenes, such as haze, sandstorm, underwater or low illumination.

HoloSR: deep learning-based super-resolution for real-time high-resolution computer-generated holograms.

This study presents HoloSR, a novel deep learning-based super-resolution approach designed to produce high-resolution computer-generated holograms from low-resolution RGBD images, enabling the real-time production of realistic three-dimensional images. The HoloSR combines the enhanced deep super-resolution network with resize and convolution layers, facilitating the direct generation of high-resolution computer-generated holograms without requiring additional interpolation. Various upscaling scales, extending up to ×4, are evaluated to assess the performance of our method. Quantitative metrics such as structural similarity and peak signal-to-noise ratio are employed to measure the quality of the reconstructed images. Our simulation and experimental results demonstrate that HoloSR successfully achieves super-resolution by generating high-resolution holograms from low-resolution RGBD inputs with supervised and unsupervised learning.

Quantifying trade-offs in satellite hardware configurations using a super-resolution framework with realistic image degradation

ABSTRACT When designing and operating Earth observation satellites, trade-offs must often be made between different hardware components, operational considerations, etc… For example, considerations include how much of the mass budget should be spent on imaging lenses?, or how long should image exposure times be?. Resulting limitations in image resolution and quality can be partially compensated for using super-resolution (SR) techniques. However, deep SR networks recently applied to satellite imagery are often trained and tested on data that lacks the typical image degrading noise present in real satellite images. In this work, we combine a method for generating realistically degraded satellite images with a deep SR network, in the context of different satellite hardware configurations and geographical types. We use this framework to assess deep SR performance given realistic remote sensing payloads across different terrain types, by evaluating payload- and terrain-dependent SR performance in reconstructing realistically degraded images. The framework allows us to model the effect of alternative satellite hardware configurations on resulting SR image quality, providing insight into optimal satellite operations and payload design in the context of SR-based image quality enhancements.

Remote Sensing Observations of a Coastal Water Environment Based on Neural Network and Spatiotemporal Fusion Technology: A Case Study of Hangzhou Bay

The coastal environment is characterized by high, multi-scale dynamics and the corresponding observations from a single remote sensing sensor are still facing challenges in achieving both high temporal and spatial resolution. This study proposed a spatiotemporal fusion model for coastal environments, which could fully enhance the efficiency of remote sensing data use and overcome the shortcomings of traditional spatiotemporal models that are insensitive to small-scale disturbances. The Enhanced Deep Super-Resolution Network (EDSR) was used to reconstruct spatial features in the lower spatial resolution GOCI-II data. The spatial features obtained instead of GOCI-II data were fed into the spatiotemporal fusion model, which enabled the fusion data to achieve an hour-by-hour observation of the water color and morphology information changes at 30 m resolution, including the changes in the spatial and temporal distributions of suspended particulate matter (SPM), the characterization of the vortex street caused by the bridge piers, the inundation process of the tidal flats, and coastline changes. In addition, this study analyzed the various factors affecting fusion accuracy, including spectral difference, errors in both temporal difference and location distance, and the structure of the EDSR model on the fusion accuracy. It is demonstrated that the location distance error and the spectral difference have the most significant impact on the fusion data, which may lead to the introduction of some ambiguous or erroneous spatial features.

Multi-image super-resolution based low complexity deep network for image compressive sensing reconstruction

Deep learning (DL) has been widely utilized in image compressive sensing (CS) to enhance the quality and speed of reconstruction. The typical deep network for CS reconstruction comprises an initial reconstruction subnetwork, followed by a cascaded deep refinement reconstruction subnetworks. This paper introduces a new low-complexity image CS deep reconstruction framework, GSRCS, which leverages multi-image based deep super-resolution technology to better address the cost constraints of practical applications. The proposed initial reconstruction module generates multiple low-resolution images in parallel by grouping the input measurements, while a high-quality, high-resolution reconstructed image is produced through a multi-image deep super-resolution network. The theoretical derivation and experimental results demonstrate that this method significantly reduces system complexity in terms of parameters and floating-point arithmetic operations, while achieving competitive reconstruction performance compared to the state-of-the-arts. Specifically, the average number of parameters is reduced by over 63%, and the computational complexity is decreased by more than 88%.

A COMPARISON OF DEEP LEARNING-BASED SUPER-RESOLUTION FRAMEWORKS FOR SENTINEL-2 IMAGERY IN URBAN AREAS

Abstract. The high-resolution images are in demand for many applications in the monitoring of urban areas. The advent of remote sensing satellites such as Sentinel-2 has made data more accessible as it provides free multispectral imagery. However, the spatial resolution of these images is not sufficient for many of the tasks. With the advent of deep learning techniques, significant progress has been made in the field of super-resolution, which has shown promising results in the improvement of the spatial resolution of satellite images. In this study, we compare four the most common deep learning-based models for the super-resolution of Sentinel-2 imagery in dense urban areas using aerial images. These methods are including enhanced deep super-resolution network (EDSR), super-resolution generative adversarial networks (ESRGAN), residual feature distillation network (RFDN), and Super-Resolution Convolutional Neural Network (SRCNN). To determine the effectiveness of the models in improving image resolution, they were evaluated using visual quality and quantitative metrics. The super-resolution results show that deep learning-based models have high potential for the generation of the high-resolution dataset from Sentinel-2 imagery in urban areas. The RFDN outperformed other deep learning-based models that achieved the peak signal-to-noise ratio (PSNR) more than 17.8.

Spatial resolution improvement by a super-resolution technique depending on training process in the background-orientated schlieren analyses

The effectiveness of the enhanced deep super-resolution network (EDSR), which is a super-resolution technique based on the deep convolutional neural network, is investigated for the improvement of spatial resolution during a background-orientated schlieren (BOS) analysis, and its suitable training process in the EDSR method is clarified. Consequently, a training dataset consisting of the simple dot patterns leads to a better image quality due to super-resolution because the image captured in the BOS measurement shows the similar dot patterns. When the image is enlarged at a large magnification ratio, it is important to adjust an image size in the training dataset individually for each magnification ratio, thereby obtaining a good estimation accuracy of the pixel displacement in the BOS analysis. A measurement error is improved by 62% compared with that of the Bicubic method, which is a classical spatial resolution improvement technique at the magnification ratio of 8. The present result shows that the EDSR method with the best training conditions provides a reasonable pixel displacement vector field up to the magnification ratio of 8 for the BOS analysis; however, its effectiveness depends on flow structure.

MR image reconstruction using iterative up and downsampling network

Magnetic resonance imaging (MRI) is one of the most significant modalities in medical imaging. It suffers from rather lengthy acquisition times, which lead to prolonged patient restriction, patient discomfort, and imaging artifacts. Hence, satisfactory MRI reconstruction from undersampled data sequences constitutes an important research problem. With the advances in deep learning (DL), a plethora of new models have been proposed to solve the MRI reconstruction problem using deep networks. On the other hand, single image super-resolution (SR) is another well-studied field that benefited from the success of DL, and it has applications in various imaging modalities. SR is the process of recovering a high-resolution image from a low-resolution image. SR models work on low-resolution images to recover missing details. MR image reconstruction on the other hand is a battle to eliminate the aliasing artifacts which originate from data downsampling. The motivation for the proposed work is based on the premise that SR approaches can possibly get adapted to MR image reconstruction. Hence in this study, inspired by the great success of deep SR networks, we customize an architecture introduced in SR setting to MRI reconstruction. This novel approach uses the iterative up and downsampling framework labeled as Iterative Up and Down Network (IUDN) for MRI reconstruction. We design two variants of the proposed network with different number of scale factors. We present extensive simulations for the proposed architecture using multiple k-space undersampling ratios. The simulation results indicate cutting edge MRI reconstruction performance for the proposed models. The networks were trained on fastMRI dataset and tested on both fastMRI and IXI datasets to show the robustness of the method. The proposed models achieved improved results in terms of PSNR and visual quality of the reconstructed image when compared to some recent state-of-the-art solutions for MRI reconstruction.

Deep learning architecture for 3D image super-resolution of late gadolinium enhanced cardiac MRI.

High-resolution late gadolinium enhanced (LGE) cardiac magnetic resonance imaging (MRI) volumes are difficult to acquire due to the limitations of the maximal breath-hold time achievable by the patient. This results in anisotropic 3D volumes of the heart with high in-plane resolution, but low-through-plane resolution. Thus, we propose a 3D convolutional neural network (CNN) approach to improve the through-plane resolution of the cardiac LGE-MRI volumes. We present a 3D CNN-based framework with two branches: a super-resolution branch to learn the mapping between low-resolution and high-resolution LGE-MRI volumes, and a gradient branch that learns the mapping between the gradient map of low-resolution LGE-MRI volumes and the gradient map of high-resolution LGE-MRI volumes. The gradient branch provides structural guidance to the CNN-based super-resolution framework. To assess the performance of the proposed CNN-based framework, we train two CNN models with and without gradient guidance, namely, dense deep back-projection network (DBPN) and enhanced deep super-resolution network. We train and evaluate our method on the 2018 atrial segmentation challenge dataset. Additionally, we also evaluate these trained models on the left atrial and scar quantification and segmentation challenge 2022 dataset to assess their generalization ability. Finally, we investigate the effect of the proposed CNN-based super-resolution framework on the 3D segmentation of the left atrium (LA) from these cardiac LGE-MRI image volumes. Experimental results demonstrate that our proposed CNN method with gradient guidance consistently outperforms bicubic interpolation and the CNN models without gradient guidance. Furthermore, the segmentation results, evaluated using Dice score, obtained using the super-resolved images generated by our proposed method are superior to the segmentation results obtained using the images generated by bicubic interpolation () and the CNN models without gradient guidance (). The presented CNN-based super-resolution method with gradient guidance improves the through-plane resolution of the LGE-MRI volumes and the structure guidance provided by the gradient branch can be useful to aid the 3D segmentation of cardiac chambers, such as LA, from the 3D LGE-MRI images.

Object recognition from enhanced underwater image using optimized deep-CNN

Object detection from underwater sea images based on deep learning techniques provides preferable results in a controlled environment. Yet, these techniques experience some challenges in detecting underwater objects due to color distortion, noise, and scattering. Hence, enhancing the underwater imaginary is important for accurately determining the objects under water. This research presents a deep learning approach for perceiving underwater objects from enhanced underwater images. Very Deep Super-Resolution Network (VDSR), which exhibits a higher visual quality, is utilized for improving the underwater image quality, thereby it is sufficient for object detection. Then, the object is detected from the enhanced underwater image through the proposed Border Collie Flamingo optimization-based deep CNN classifier (BCFO-based deep CNN). The developed BCFO-based algorithm is the main highlight of the research, which effectively tunes the classifier’s hyperparameter. The evaluation is established using the UIEB and DUO datasets on the basis of performance standards, such as specificity, accuracy, and sensitivity. When the training percentage is 80 and the [Formula: see text]-fold is 10, the suggested model achieved accuracy results of 93.89% and 95.24%, sensitivity results of 95.93 and 97.29%, and specificity results of 98.64% and 99%, which is very efficient compared to some existing approaches.

Prostatic urinary tract visualization with super-resolution deep learning models.

In urethra-sparing radiation therapy, prostatic urinary tract visualization is important in decreasing the urinary side effect. A methodology has been developed to visualize the prostatic urinary tract using post-urination magnetic resonance imaging (PU-MRI) without a urethral catheter. This study investigated whether the combination of PU-MRI and super-resolution (SR) deep learning models improves the visibility of the prostatic urinary tract. We enrolled 30 patients who had previously undergone real-time-image-gated spot scanning proton therapy by insertion of fiducial markers. PU-MRI was performed using a non-contrast high-resolution two-dimensional T2-weighted turbo spin-echo imaging sequence. Four different SR deep learning models were used: the enhanced deep SR network (EDSR), widely activated SR network (WDSR), SR generative adversarial network (SRGAN), and residual dense network (RDN). The complex wavelet structural similarity index measure (CW-SSIM) was used to quantitatively assess the performance of the proposed SR images compared to PU-MRI. Two radiation oncologists used a 1-to-5 scale to subjectively evaluate the visibility of the prostatic urinary tract. Cohen's weighted kappa (k) was used as a measure of agreement of inter-operator reliability. The mean CW-SSIM in EDSR, WDSR, SRGAN, and RDN was 99.86%, 99.89%, 99.30%, and 99.67%, respectively. The mean prostatic urinary tract visibility scores of the radiation oncologists were 3.70 and 3.53 for PU-MRI (k = 0.93), 3.67 and 2.70 for EDSR (k = 0.89), 3.70 and 2.73 for WDSR (k = 0.88), 3.67 and 2.73 for SRGAN (k = 0.88), and 4.37 and 3.73 for RDN (k = 0.93), respectively. The results suggest that SR images using RDN are similar to the original images, and the SR deep learning models subjectively improve the visibility of the prostatic urinary tract.

The Best of Both Worlds: A Framework for Combining Degradation Prediction with High Performance Super-Resolution Networks.

To date, the best-performing blind super-resolution (SR) techniques follow one of two paradigms: (A) train standard SR networks on synthetic low-resolution-high-resolution (LR-HR) pairs or (B) predict the degradations of an LR image and then use these to inform a customised SR network. Despite significant progress, subscribers to the former miss out on useful degradation information and followers of the latter rely on weaker SR networks, which are significantly outperformed by the latest architectural advancements. In this work, we present a framework for combining any blind SR prediction mechanism with any deep SR network. We show that a single lightweight metadata insertion block together with a degradation prediction mechanism can allow non-blind SR architectures to rival or outperform state-of-the-art dedicated blind SR networks. We implement various contrastive and iterative degradation prediction schemes and show they are readily compatible with high-performance SR networks such as RCAN and HAN within our framework. Furthermore, we demonstrate our framework's robustness by successfully performing blind SR on images degraded with blurring, noise and compression. This represents the first explicit combined blind prediction and SR of images degraded with such a complex pipeline, acting as a baseline for further advancements.

Lightweight Dual Mutual-Feedback Network for Artificial Intelligence in Medical Image Super-Resolution

As a result of hardware resource constraints, it is difficult to obtain medical images with a sufficient resolution to diagnose small lesions. Recently, super-resolution (SR) was introduced into the field of medicine to enhance and restore medical image details so as to help doctors make more accurate diagnoses of lesions. High-frequency information enhances the accuracy of the image reconstruction, which is demonstrated by deep SR networks. However, deep networks are not applicable to resource-constrained medical devices because they have too many parameters, which requires a lot of memory and higher processor computing power. For this reason, a lightweight SR network that demonstrates good performance is needed to improve the resolution of medical images. A feedback mechanism enables the previous layers to perceive high-frequency information of the latter layers, but no new parameters are introduced, which is rarely used in lightweight networks. Therefore, in this work, a lightweight dual mutual-feedback network (DMFN) is proposed for medical image super-resolution, which contains two back-projection units that operate in a dual mutual-feedback manner. The features generated by the up-projection unit are fed back into the down-projection unit and, simultaneously, the features generated by the down-projection unit are fed back into the up-projection unit. Moreover, a contrast-enhanced residual block (CRB) is proposed as each cell block used in projection units, which enhances the pixel contrast in the channel and spatial dimensions. Finally, we designed a unity feedback to down-sample the SR result as the inverse process of SR. Furthermore, we compared it with the input LR to narrow the solution space of the SR function. The final ablation studies and comparison results show that our DMFN performs well without utilizing a large amount of computing resources. Thus, it can be used in resource-constrained medical devices to obtain medical images with better resolutions.

A database of students’ spontaneous actions in the real classroom environment

Students’ actions in the classroom play a key role in their studying performance in class. This study focuses on the automatic recognition of students’ spontaneous actions in a real classroom environment. A database of students’ spontaneous actions is first established in this study. Our database consists of 4917 images of students’ spontaneous actions, including the actions of students in 10 learning states. Based on the characteristics of students’ actions in the classroom, we proposed a new 11-layer convolutional neural network algorithm based on an enhanced deep super resolution network. In this algorithm, a data augmentation method is introduced for data processing. The experimental results show that the proposed algorithm effectively improved the accuracy of action recognition on the database established in this paper. Such a spontaneous database proposed in this paper provides good data support for the study of students’ actions in the classroom.

Super-resolution reconstruction of pneumocystis carinii pneumonia images based on generative confrontation network

ObjectivePneumocystis carinii pneumonia, also known as pneumocystis carinii pneumonia (PCP), is an interstitial plasma cell pneumonia caused by pneumocystis spp. It is a conditional lung infectious disease. Because the early and correct diagnosis of PCP has a great influence on the prognosis of patients, the image processing of PCP's high-resolution CT (HRCT) is extremely important. Traditional image super-resolution reconstruction algorithms have difficulties in network training and artifacts in generated images. The super-resolution reconstruction algorithm of generative counter-networks can optimize these two problems well. MethodsIn this paper, the texture enhanced super-resolution generative adversarial network (TESRGAN) is based on a generative confrontation network, which mainly includes a generative network and a discriminant network. In order to improve the quality of image reconstruction, TESRGAN improved the structure of the Super-Resolution Generative Adversarial Network (SRGAN) generation network, removed all BN layers in SRGAN, and replaced the ReLU function with the LeakyReLU function as the nonlinear activation function of the network to avoid the disappearance of the gradient. Experimental resultsThe TESRGAN algorithm in this paper is compared with the image reconstruction results of Bicubic, SRGAN, Enhanced Deep Super-Resolution network (EDSR), and ESRGAN. Compared with algorithms such as SRGAN and EDSR, our algorithm has clearer texture details and more accurate brightness information without extending the running time. Our reconstruction algorithm can improve the accuracy of image low-frequency information. ConclusionThe texture details of the reconstruction result are clearer and the brightness information is more accurate, which is more in line with the requirements of visual sensory evaluation.

MRI super-resolution via realistic downsampling with adversarial learning

Many deep learning (DL) frameworks have demonstrated state-of-the-art performance in the super-resolution (SR) task of magnetic resonance imaging, but most performances have been achieved with simulated low-resolution (LR) images rather than LR images from real acquisition. Due to the limited generalizability of the SR network, enhancement is not guaranteed for real LR images because of the unreality of the training LR images. In this study, we proposed a DL-based SR framework with an emphasis on data construction to achieve better performance on real LR MR images. The framework comprised two steps: (a) downsampling training using a generative adversarial network (GAN) to construct more realistic and perfectly matched LR/high-resolution (HR) pairs. The downsampling GAN input was real LR and HR images. The generator translated the HR images to LR images and the discriminator distinguished the patch-level difference between the synthetic and real LR images. (b) SR training was performed using an enhance4d deep super-resolution network (EDSR). In the controlled experiments, three EDSRs were trained using our proposed method, Gaussian blur, and k-space zero-filling. As for the data, liver MR images were obtained from 24 patients using breath-hold serial LR and HR scans (only HR images were used in the conventional methods). The k-space zero-filling group delivered almost zero enhancement on the real LR images and the Gaussian group produced a considerable number of artifacts. The proposed method exhibited significantly better resolution enhancement and fewer artifacts compared with the other two networks. Our method outperformed the Gaussian method by an improvement of 0.111 ± 0.016 in the structural similarity index and 2.76 ± 0.98 dB in the peak signal-to-noise ratio. The blind/reference-less image spatial quality evaluator metric of the conventional Gaussian method and proposed method were 46.6 ± 4.2 and 34.1 ± 2.4, respectively.

Imaging depth adaptive resolution enhancement for optical coherence tomography via deep neural network with external attention

Optical coherence tomography (OCT) is a promising non-invasive imaging technique that owns many biomedical applications. In this paper, a deep neural network is proposed for enhancing the spatial resolution of OCT en face images. Different from the previous reports, the proposed can recover high-resolution en face images from low-resolution en face images at arbitrary imaging depth. This kind of imaging depth adaptive resolution enhancement is achieved through an external attention mechanism, which takes advantage of morphological similarity between the arbitrary-depth and full-depth en face images. Firstly, the deep feature maps are extracted by a feature extraction network from the arbitrary-depth and full-depth en face images. Secondly, the morphological similarity between the deep feature maps is extracted and utilized to emphasize the features strongly correlated to the vessel structures by using the external attention network. Finally, the SR image is recovered from the enhanced feature map through an up-sampling network. The proposed network is tested on a clinical skin OCT data set and an open-access retinal OCT dataset. The results show that the proposed external attention mechanism can suppress invalid features and enhance significant features in our tasks. For all tests, the proposed SR network outperformed the traditional image interpolation method, e.g. bi-cubic method, and the state-of-the-art image super-resolution networks, e.g. enhanced deep super-resolution network, residual channel attention network, and second-order attention network. The proposed method may increase the quantitative clinical assessment of micro-vascular diseases which is limited by OCT imaging device resolution.

Multiple Optimizations-Based ESRFBN Super-Resolution Network Algorithm for MR Images

Magnetic resonance (MR) images can detect small pathological tissue with the size of 3–5 image pixels at an early stage, which is of great significance in the localization of pathological lesions and the diagnosis of disease. High-resolution MR images can provide clearer structural details and help doctors to analyze and diagnose the disease correctly. In this paper, MR super-resolution based on the multiple optimizations-based Enhanced Super Resolution Feed Back Network (ESRFBN) is proposed. The method realizes network optimization from the three perspectives of network structure, data characteristics and heterogeneous network integration. Firstly, a super-resolution network structure based on multi-slice input optimization is proposed to make full use of the structural similarity between samples. Secondly, aiming at the problem that the L1 or L2 loss function is based on a per-pixel comparison of differences, without considering human visual perception, the optimization method of multiple loss function cascade is proposed, which combines the L1 loss function to retain the color and brightness characteristics and the MS-SSIM loss function to retain the contrast characteristics of the high-frequency region better, so that the depth model has better characterization performance; thirdly, in view of the problem that large deep learning networks are difficult to balance model complexity and training difficulty, a heterogeneous network fusion method is proposed. For multiple independent deep super-resolution networks, the output of a single network is integrated through an additional fusion layer, which broadens the width of the network, and can effectively improve the mapping and characterization capabilities of high- and low-resolution features. The experimental results on two super-resolution scales and on MR images datasets of four human body parts show that the proposed large-sample space learning super-resolution method effectively improves the super-resolution performance.