Glnet For Target Detection In Millimeter Wave Images

Millimeter wave imaging can be used to detect objects hidden under clothes and it does no harm to human body, thus it is useful for human body inspection. However, due to limitation of technology, the image is usually noisy, which leads to a great challenge for target detection without human supervision. Based on CNN (convolutional neural network), this paper proposes a Global and Local Network, namely GLNet. The GLNet consists of two convolution neural networks: a global faster and coarse region proposal network (GNet) aiming to extract potential regions containing targets and a local fine network (LNet) for further classification and position calibration. The final results combining the outputs of GNet and LNet shows a better performance than other method. The proposed model achieves 73.5% AP on target detection in our millimeter wave images dataset while keeps 35 frames per seconds for real time detection.