Research And Implementation Of Target Detection Method For SAR Image

The Synthetic Aperture Radar(SAR)is an imaging Radar with the characteristics of all-time,all-weather,high-resolution,wide-width and so on.SAR image target detection and recognition,as the core part of SAR image processing technology,has a very important application value in detecting and identifying target objects and extracting effective information in complex scenes.Nowadays,deep learning methods are very popular in the field of optical image processing,and many excellent algorithms and achievements have been accumulated.Thesis combined with the deep learning method,studies the target detection and recognition technology of SAR images,improves the construction method of training data and puts forward the optimization scheme of the algorithm according to the particularity of SAR images.The main contents of this thesis can be summarized as follows:Firstly,the target detection algorithm based on candidate regions in deep learning is studied in this part.According to the particularity of SAR image and the difference from optical image,this thesis improves the traditional Faster R-CNN algorithm and proposes the Faster R-CNN Optimization algorithm.By comparing the experimental data and analyzing the experimental results,it is found that the Faster R-CNN Optimization network can improve the experimental effect of the SAR image target detection task by using different backbone feature extraction network Res Net-50 by horizontally changing,while vertical adjustment uses different hyperparameters for SAR image target detection task performance and effect also have a certain impact.Secondly,a polynomial regression optimization algorithm is designed for the task of SAR image target detection and recognition,based on the traditional hyperparameter optimization method and regression analysis in statistics.The polynomial regression optimization algorithm establishes a regression model by optimizing the data,which can quickly screen out the optimal hyperparameters and improve the efficiency of SAR image target detection and recognition.Thirdly,according to the characteristics of SAR image data,experiments are designed to classify the SAR image data set according to the sea state and the scene,and train and test the model respectively.The results show that the data set is divided according to different sea conditions and scenes,and the targeted training of the model is helpful to improve the effect of the model for target detection and recognition in SAR images under specific sea conditions and scenes.Fourthly,a target detection and recognition system based on SAR images is designed by combining the target detection and recognition model with the polynomial regression optimization algorithm.The system includes a data management module,a model management module and a test analysis module.By using this system,the target detection and recognition model can be trained and optimized,the target detection model can be visualized and analyzed,and the polynomial regression optimization algorithm can be used to efficiently complete the SAR image target detection and recognition task.According to the actual characteristics of SAR images,this thesis makes corresponding improvements to the deep learning method of optical image processing,which not only improves the efficiency of SAR image target detection,but also improves the effect of SAR image target detection in specific environments.