GPU parallel implementation and optimisation of SAR target recognition method

Abstract

The SAR target recognition based on optimised GPU parallel algorithm is proposed here. In general, with the rapid increment of the data dimension and the amount of data of SAR images, the traditional CPU-based target recognition algorithm cannot meet the requirements of real-time processing. Here, the target recognition algorithm which includes feature extraction and the classification is investigated and then parallel decomposed and optimised. First, the algorithms are investigated and parallel decomposed, including the principal component analysis, linear discriminant analysis, and non-negative matrix factorisation feature extraction technologies, and the support vector machines classifier. Then, the three feature extraction methods and sequential minimal optimisation algorithm are realised. Finally, the causes of compute unified device architecture programme running speed in target recognition algorithm are deeply analysed, and the algorithm is optimised from three aspects: communication, access, and instruction flow. According to the experiments, the optimised GPU-based parallel implementation of the target recognition algorithm has been optimised to obtain about 25–30 times performance upgrade

Keywords

• support vector machines
• graphics processing units
• principal component analysis
• parallel architectures
• synthetic aperture radar
• parallel algorithms
• pattern classification
• matrix decomposition
• optimisation
• radar imaging
• feature extraction
• sar target recognition method
• optimised gpu parallel algorithm
• sar images
• traditional cpu-based target recognition algorithm
• nonnegative matrix factorisation feature extraction technologies
• feature extraction methods
• sequential minimal optimisation algorithm
• optimised gpu-based parallel implementation