Super-resolution reconstructed convolution neural network (SRCNN) is widely used in image quality improvement of single image. Traditional SRCNN training uses the loss function of minimum mean square error (MSE) and the method based on stochastic gradient descent (SGD) to optimize. Its learning rate adjustment strategy is limited by pre-specified adjustment rules, and it is difficult to select the initial value. Considering the complex texture and low resolution of remote sensing images, a deconvolution layer is proposed to replace the bi-cubic interpolation enlarged image in the traditional SRCNN network to overcome the mosaic effect. At the same time, Adam optimizer is used to control the network training. After considering the first and second moment estimation of gradient comprehensively, the update step is calculated. Thus, the adaptive update of learning rate is realized and the speed of network training is greatly accelerated. The simulation results show that this method has advantages in edge reconstruction and texture details compared with the conventional super-resolution reconstruction algorithm.
During space reconnaissance applications, edge detection from remote sensing imagery plays an important role in the target recognition processing. However, traditional edge detection methods usually only utilize the high-frequency information in one image. Since low-frequency elements may be aliasing with high-frequency parts, the edges extracted may be unconnected under complex topography, different objects and imaging conditions. This paper proposes a novel image edge detection method based on Non-Subsampled Contourlet Transform (NSCT) to keep the object boundary continuously. It transforms the image into Contourlet domain in both high-frequency and low-frequency sub-bands respectively. Depending on the feature of flexible directivity reservation of an image during NSCT, the further edge extraction consists of 3 steps: firstly, the elements of the high-frequency coefficient matrix in Contourlet domain are filtered with high values left using adaptive thresholds. Then the low-frequency edge information is extracted via Canny operator from the low-frequency sub-band information. Finally, to achieve a more consistent edge image, the low-frequency edge image is achieved according to the low-frequency matrix and adopted to compensate the high-frequency image with the isolated noise points eliminated as well. The numerical simulation and practical test results show the higher effectiveness and robustness of the proposed algorithm when comparing with the classical edge detectors, such as Sobel operator, Canny operator, Log operator and Prewitt operator, etc.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.