Analysis and Recognition of Heart Sound Based on NCS2 Neural Computing Stick

Abstract

At present, the recognition and analysis of heart sound signal was usually run by using high-performance PC. It was hardly done by embedded devices due to limited resources. Provide a portable device for assisting in the initial diagnosis of congenital heart disease (CHD) for doctors with outdated equipment in remote mountainous areas. A novel embedded heart sound analysis and recognition system based on Raspberry pi 3b+ with a NCS2 neural computing stick was put forward in this paper. Firstly, the OpenVINO software platform launched by Intel was used to transfer the ssd_inception_v2 model into the Raspberry Pi after performing transfer learning optimization. Then, reasoning calculation was carried out in Raspberry pi with neural computing stick. Neural computing stick is a deep learning and reasoning tool based on USB mode and an independent artificial intelligence accelerator. NCS2 neural computing stick was used to realize the heart sound analysis and recognition of embedded devices. The sensitivity of the experimental results is 80.7%, the specificity is 95.5%, and the accuracy is 91.4%. The experimental results show that the system has the advantages of low power dissipation, low cost, small size, fast speed, and high recognition rate. It can be used for machine assisted diagnosis of congenital heart disease.