An Adaptive Feature Learning Model for Sequential Radar High Resolution Range Profile Recognition.

Area of Science:

• Radar Signal Processing
• Machine Learning
• Pattern Recognition

Background:

• Radar High Resolution Range Profiles (HRRPs) are crucial for target recognition.
• Traditional methods like Hidden Markov Models (HMMs) require extensive preprocessing and struggle with global pattern capture.

Purpose of the Study:

• To introduce a novel, end-to-end feature learning method for HRRP sequence recognition.
• To develop a model capable of capturing global patterns and adaptively learning features from raw HRRP data.

Main Methods:

• A Discriminative Infinite Restricted Boltzmann Machine (Dis-iRBM) is proposed to jointly model and discriminate HRRP sequences.
• The Dis-iRBM processes raw HRRP sequences directly, unlike HMMs that require signal preprocessing.

Main Results:

• The Dis-iRBM effectively captures global patterns within HRRP sequences, overcoming the local dynamics limitation of HMMs.
• The model adaptively learns features based on HRRP complexity and sequence length, demonstrating robustness.
• Experiments on the Moving and Stationary Target Acquisition and Recognition (MSTAR) database show superior efficiency and robustness.

Conclusions:

• The Dis-iRBM presents a significant advancement in HRRP sequence recognition, offering an end-to-end, feature-adaptive solution.
• This method provides a more effective approach to radar target recognition compared to existing HMM-based techniques.