基于改进RPN的Faster-RCNN网络SAR图像车辆目标检测方法

Method for vehicle target detection on SAR image based on improved RPN in Faster-RCNN

针对传统Faster-RCNN方法中候选区域生成网络(RPN)模块在进行目标检测时对目标特征提取不够充分的问题,提出一种基于改进RPN的Faster-RCNN网络SAR图像车辆目标检测方法.首先基于VGG-16网络提取出图片的多层特征,然后利用卷积核对最深的3个特征层作进一步的特征提取和正则化处理,最后对处理后的3个特征层进行信息融合.利用MSTAR数据集中车辆目标SAR图像和自然背景SAR图像,通过图像分割和贴图的方式制作了SAR场景数据集,对所改进网络进行训练和测试.实验结果表明,在SAR图像车辆目标检测中,与传统RPN相比,改进RPN收敛速度更快,不仅将检测结果的查准率从97.7%提高到了99.7%,虚警率明显降低,而且泛化性能更强,针对训练范围以外的目标,能将查准率由98.0%提高到99.0%.

Aiming at the problem that the region proposal network(RPN)module couldn't adequately extract target features when performing target detection in the traditional Faster-RCNN method,a method for vehicle target detection on the SAR image based on the improved RPN in Faster-RCNN was proposed.First,the multi-layer features of the image were extracted based on the VGG-16 network.Then the deepest three feature layers were further extracted and regularized using convolution kernels.Finally,the information fusion was performed on the three processed feature layers.Using the vehicle target SAR image and the natural background SAR image in the MSTAR data set,the SAR scene data set was created by image segmentation and texture,and the improved network was trained and tested.Experimental results show that in the detection of the vehicle targets on SAR images,compared with the traditional RPN,the improved RPN has a faster convergence speed,improving the accuracy of the test results from 97.7%to 99.7%,with a lower false alarm rate,and has stronger generalization performance.For targets outside the training range,the accuracy can be increased from 98.0%to 99.0%.