Foliage-concealed Target Detection Based on Deep Learning in Low Frequency UWB SAR Images

Low frequency ultra-wideband synthetic aperture radar (UWB SAR) has the well ability of the foliage penetration and high-resolution imaging, which can detect foliage-concealed targets in the forests. However, the low frequency UWB SAR images often contain many non-target strong scattering points like thick tree trunks. This paper proposes a foliage-concealed target detection method based on the deep learning in the low frequency UWB SAR images. First, based on the CARABAS-II SAR dataset, the low frequency UWB SAR dataset of foliage-concealed target and corresponding annotation is generated by random cutting and data augmentation. Then, a convolutional neural network (CNN) is designed for prediction of the foliage-concealed target, and the output is a predicted image of the same size as the input image. Besides, the generated dataset is used to the designed CNN, and the training is stop when the network converges. Finally, based on the characteristics of the network output result, the variance filtering and binarization operations are used for the post-processing of the network output image, to obtain the final prediction result. Experimental results show that the proposed method can realize the foliage-concealed target detection in low frequency UWB SAR images and ensure the rapid detection of multiple targets with the high accuracy.