基于YOLO神经网络的雷达目标成像识别评估研究

针对复杂地面背景环境下的武器装备精确探测识别需求,采用Lee增强滤波、对比度自适应直方图均衡化和能量归一化等图像预处理方法,提高SAR图像质量;通过引入两个可学习的参数和采用基于非极大值抑制(NMS)方法构建了优化的YOLO神经网络目标识别方法,对基于轮廓、纹理等特征的地面目标SAR图像自动识别进行了实验.实验结果表明...

Research and evaluation on radar target imaging recognition based on YOLO neural network

Aiming at the requirements of accurate detection and identification of weapons and equipment under complex ground background,this paper uses image preprocessing methods such as Lee enhancement filtering,contrast adaptive histogram equalization,energy normalization,etc.to improve the quality of SAR images.And then,the paper brings in two learnable parameters and adopts the method based on non-maximum suppression to construct the optimized YOLO neural network target recognition method.Finally,experiments of the automatic SAR image recognition of ground targets based on contour and texture features are carried out.The experimental results show that compared with the deformation convolutional neural networks(DPM)and the region-based convolutional neural networks(RCNN),the target recognition rate of the optimized YOLO network is increased by more than 10%.This provides a way to evaluate the stealth performance based on target imaging recognition.