With the rapid development of artificial intelligence technology and the emergence of a large number of innovative theories, the concept of deep learning is widely used in object detection, speech recognition, language translation and other fields. One of the important practices is target recognition in SAR images. Although it shows certain effectiveness in some researches, when using deep learning algorithm, there are still many problems that have not yet been solved. For example, the convergence of algorithms has not shown intuitively, although the high precision of experimental results can be obtained. There are many reasons that lead to the results divergence, such as the size of the database, the type of model, and the algorithm used in the experiment. This paper aims at analyzing the factors that influence the convergence of the results from the perspective of the CNN algorithm. The goal can be achieved by means of constraint of convergence condition. Firstly, by controlling the amount of data in the database, the influence of the size of the database will be determined. Secondly, the radius of convergence will be analyzed, based on which, the scope of application will be found. Combining the above two factors, a corresponding method can be given in the final paper, which gives rise to the convergence of the result. Finally, the correctness of the above theories will be explained by conducting experiments using the MSTAR database.
With the rapid development of sensor and communication technology, the volume and the resolution of the data became increasingly high. Compressive Sensing theory allows signal compressed at a rate much lower than the Nyquist rate, which is promising to deal with big data acquisition and transmission. Compressive sensing has been applied in a variety of fields such as clutter suppression, image/video reconstruction, and real time processing. Most of the conventional algorithms for the estimation of the original signal, for instance, Total Variation (TV), consist of consistency error and constraint terms, the latter of which is quite influential on the quality of reconstructed image. The results subject to different constraints may vary greatly, for example, the conventional TV constraint suffers from the step effect, while the Higher Degree Total Variation (HDTV) may have the defect of edge blur. Besides, the computational cost is another problem, which needs to be considered. In this paper, a constraint refinement based algorithm for compressive sensing image reconstruction is proposed. Firstly, the construction of the constraint term is studied. For images that show different characteristics (for example, the richness of texture, etc.), the appropriate constraints for different cases are discussed. Secondly, a modified constraint is introduced to overcome the defect of the aforementioned algorithms. Moreover, a fast approximation algorithm to enhance the calculation efficiency is proposed based on the introducing of an auxiliary function to cross update. The visual and quantitative assessment both prove the superiority of the proposed constraint refinement method in terms of SNR, SSIM, and PSNR.
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.