Study On SAR Image Target Recognition

Synthetic Aperture Radar(SAR)is a radar mode that can obtain high resolution imaging all day and all weather.Its development embodies the progress of modern radar technology.With the continuous improvement of imaging equipment and technology,SAR imaging resolution and imaging quality have been significantly improved and have been widely used in multi-layer fields such as geographic exploration,ocean observation,disaster prediction,military reconnaissance,etc.SAR image target recognition bears the task of fast and accurate interpretation of massive image data.With the improvement of theories such as machine learning and deep learning,the research direction has become a hot topic of common concern and discussion among scholars.This paper mainly studies and improves from the following three aspects:1.A kind of image segmentation algorithm based on Improved Shuffled Frog Leaping Algorithm(ISFLA)and Fuzzy C-means(FCM)Clustering is proposed.FCM integrates the essence of fuzzy theory,and can obtain more flexible clustering results than traditional hard clustering algorithms by continuously updating clustering centers and membership functions.The calculation method of updating step size in the original Shuffled Frog Leaping Algorithm(SFLA)is improved so that the algorithm is balanced in local and global search.ISFLA and FCM are combined to solve the defect which FCM is sensitive to initial values and easily limited to local optimization.ISFLA-FCM algorithm is applied to SAR image segmentation,which can achieve lower error rate and better image segmentation results.2.A kind of SAR image target recognition algorithm based on Support Vector Machine(SVM)optimized by Sine Cosine Algorithm(SCA)is proposed.Different from traditional SAR image feature extraction methods such as Hu invariant moments,affine invariant moments and moment of inertia,the fused Gabor wavelet features and Histogram of Oriented Gradients(HOG)features are used as SAR image feature vectors.SCA is used to optimize the penalty coefficient and kernel function coefficient of SVM model,and the optimized SVM model is used for training and testing.The simulation results show that the SAR image target recognition algorithm based on SCASVM can obtain higher recognition accuracy than the traditional target recognition algorithms.3.A kind of SAR image target recognition algorithm based on Extreme Gradient Boosting(XGBoost)and Convolutional Neural Networks(CNN)is proposed.Firstly,the SAR image data set is augmented by image processing such as rotation,rollover,stretching,brightness enhancement,Gaussian blur,etc;Then,the CNN structure is optimized by using swish activation function,adding deactivation layer and Adam optimizer;Finally,XGBoost is used to replace the Softmax layer in CNN model for classification and identification at the top layer of network model.The simulation results show that the SAR image target recognition algorithm based on CNN-XGBoost uses CNN to extract image features efficiently,and XGBoost is used as the top-level identifier of the network,which effectively improves the accuracy of SAR image target recognition.