基于去模糊空间变换RCNN的毫米波图像目标检测（英文）

提出一种包含去模糊的空间变换区域卷积神经网络的目标检测算法.首先,基于主动毫米波圆柱扫描成像原理对人体进行三维成像(频率24~30 GHz),建立毫米波图像数据集.然后,估计毫米波图像的模糊核,通过卷积去噪网络获得图像先验知识,将其集成到半二次分裂的优化方法中,以实现非盲目去模糊.最后,由定位网络、网格生成器和采样网络三部分组成空间变换网络,将它融入到特征提取网络中,在去模糊后实现目标检测.通过该非盲目去模糊算法得到的图像的峰值信噪比可达27.49 dB,目标检测算法的平均精度可达80.9%.实验结果表明,与现有的先进方法相比,该方法可以有效地提高图像质量和检测精度,为毫米波图像中隐藏危险品的目标检测提供了新的技术支持.

Object Detection of Millimeter-wave Image Based on Spatial-transformer RCNN with Deblurring

An object detection algorithm of spatial-transformer regional convolutional neural network with deblurring was proposed.Firstly,based on the principle of active millimeter-wave cylindrical scanning imaging,the human body is three-dimensionally imaged(frequency range from 24 GHz to 30 GHz),and a millimeter wave image data set is established.Then the blur kernel of the millimeter-wave image is estimated.The image prior knowledge is obtained by the convolutional denoiser network and is integrated into an optimization method of half quadratic splitting to achieve non-blind deblurring.Finally,the spatial transform network,composed of a localization net,a grid generator,and a sampling network,is inserted into the feature extraction network to achieve object detection after deblurring.With the proposed non-blind deblurring algorithm,peak signal to noise ratio of the image can reach 27.49 dB.Mean average precision of object detection algorithm can reach 80.9%.The experimental results show that the image quality and detection accuracy can effectively be improved through the proposed method compared with some state-of-the-art methods.New technical support is provided for object detection of hidden dangerous goods in millimeter-wave images.