Automated Optimization for FPGA-Based Sonar Object Recognition Systems

The sonar system has been widely adopted in underwater environment exploration. The system generally needs to first reconstruct sonar images and then perform object recognition of sonar images. Current designs often focus on either sonar image reconstruction, or object recognition separately, lacking an integrated approach. The changing conditions and diverse tasks have necessitated rapid trade-offs among enormous design factors to attain short processing latency and low power consumption. In this paper, we proposes an automated methodology to effectively explore the design space and optimize the sonar system. The generated design is synthesized on FPGAs. Our experiments have shown that the proposed design achieves 3.24 x performance improvement over manual designs.