A polarization HRRP target classification method based on one-dimensional convolutional attention neural network

High resolution range profile (HRRP) signal can represent the structure information of observed targets, and it can be easy obtained. And for the polarization HRRP signals, the polarization information can further display the symmetry, spatial orientation and physical property. Using HRRP signals to classify targets has been widespread concern and has been applied in military and civilian fields. Traditional classification methods always extract handmade features, and then choose a classifier for target classification. With the extensive performance improvement of neural network in fields, this paper introduces the neural network into HRRP classification and proposes a polarization HRRP target classification method based on one-dimensional convolutional attention neural network. Considering the data representation on HRRP, the proposed method employs one-dimensional neural network operators. The network backbone is composed of spindle-shaped residual blocks. And in the spindle-shaped residual blocks, an attention module is added to enhance the target representation. Experiment results show that the proposed method obtains more high classification accuracy than traditional methods in multiple datasets.