recognized oil rigs from Sentinel-1 SAR images using VGG-16 and VGG-19. The method in can settle the over-fitting problem of CNN caused by insufficient training data to a certain extent.
constructed a new network called A-ConvNet for SAR ATR, which has no fully connected layers and only contains convolution layers. Currently, there are many SAR ATR methods based on CNN. Several existing studies show that the deep features learned through convolution operations inclined to be more discriminatory for different types of targets. Convolutional neural network (CNN) is a typical network used in image classification, which achieves feature extraction and classification using a unified framework and can automatically learn features more suitable for classification. In order to imitate the human brain’s cognitive mechanism, deep learning constructs a multi-level model structure to achieve multi-layer nonlinear transformation, through which the original data can be mapped into a feature space more suitable for recognition. With the emergence of extensive data sets and the growth of computer processing power, deep learning has been widely used in many fields. The experimental results on the Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset prove the capability of the SAR ATR method proposed in this letter. Finally, the high-level features obtained by the two branches are fused to recognize the target. There are two branches in the proposed network, one extracts the more discriminative image features from the input SAR image the other extracts physically meaningful features from the ASC schematic map that reflects the local structure of the target corresponding to each ASC. Therefore, we propose a network to comprehensively use the image features and the features related to ASCs for improving the performance of SAR ATR. For SAR images, attributed scattering centers (ASCs) reflect the electromagnetic scattering characteristics and the local structures of the target, which are useful for SAR ATR. However, most of the SAR ATR methods using CNN mainly use the image features of SAR images and make little use of the unique electromagnetic scattering characteristics of SAR images. It is very common to apply convolutional neural networks (CNNs) to synthetic aperture radar (SAR) automatic target recognition (ATR).
0 kommentar(er)
