Research On SAR Target Recognition Algorithm Based On CNN

SAR image target recognition is to extract features from the target SAR image and determine the category attributes of the target.It has a wide range of military and civil values.It can be applied to battlefield monitoring,guided attack,strike effect evaluation,marine resource detection,environmental and geomorphic detection and natural disaster evaluation.It has become the hot issue domestically and abroad.Convolutional neural networks(CNN)is an important deep learning model,including multi-layer convolution layer and pooling layer,has successfully imitated the perception process of biological neurons,especially suitable for intelligent image interpretation,and has played a good role in image target detection and recognition in the optical field.Therefore,the study of SAR target recognition technology based on convolutional neural network is of great significance for the development of SAR ATR technology.In this paper,focusing on the problem of SAR Target Recognition Based on convolutional neural network,the methods of SAR target recognition algorithm based on CNN under small sample data set and the integrated algorithm of SAR target detection and recognition based on the single shot multibox detector(SSD)are studied.The main contents are as follows:1.Aiming at the problem of convolutional neural network over fitting caused by small sample SAR training samples,the model of generative adversarial networks(GAN)is studied.Expanding the original SAR image set by using the generation countermeasure network model,alleviating the reduction of the generalization ability of the model caused by the lack of training samples.2.Aiming at the problem of poor recognition performance of CNN model under the extended conditions such as the change of SAR target configuration and the large change of imaging pitch angle,an improved pooled CNN model is established,which can effectively improve the recognition performance of CNN model under various SAR extended conditions without significantly improving the algorithm complexity.3.Aiming at the problem that the conventional SSD target detection and recognition algorithm can not automatically set the default frame size,this paper proposes an algorithm based on kmeans to set the size of the default box.By clustering the training sample data with kmeans algorithm,the default frame size setting for specific target recognition problem is automatically completed,and the recognition performance of the algorithm is improved4.Aiming at the problem of poor performance of conventional SSD target detection and recognition algorithm in SAR image,a feature fusion based SSD network model structure is constructed.The model structure abandons the traditional SSD algorithm,which only uses high-level feature map,and generates auxiliary feature map by feature fusion,which effectively improves the recognition accuracy and speed in SAR image.The above contents are verified by simulation experiments,which can achieve accurate and efficient SAR image target recognition.