Design of human-like behavior learning machine based on neural network FPGA

• Published in: Journal of physics. Conference series Vol. 1423; no. 1; pp. 12062 - 12067
• Main Authors: Tang, Ming, Hao, Shengli
• Format: Journal Article
• Language: English
• Published: Bristol IOP Publishing 01.12.2019
• Subjects: Algorithms; Bionics; Design; Field programmable gate arrays; Image processing; Image reconstruction; Knowledge bases (artificial intelligence); Machine learning; Modules; Neural networks; Object recognition; Parallax; Physics; Power consumption; Systems design; Underwater exploration
• ISSN: 1742-6588; 1742-6596
• DOI: 10.1088/1742-6596/1423/1/012062

Neural network has been widely used in image recognition, because it can imitate the behavior of biological visual nerve to obtain high recognition accuracy. Aiming at image recognition, this paper studies the machine self-learning system of image recognition results based on neural network FPGA, designs the data sensing and image recognition processing module based on neural network, and studies the key problems of the system self-recognition and machine self-learning process. The image recognition module uses bionics and parallax principle to get the synchronous exposure image of surrounding scenery, reconstruct the three-dimensional shape and position of surrounding objects, and get the information needed by the system design through image processing algorithm. On the basis of this module, a machine self-learning algorithm based on the result of image recognition is designed to build a knowledge base for the machine and guide the machine by modifying the weight and making decision to realize human-like behavior of machine self-learning. The experimental results show that the recognition accuracy of dynamic image recognition in image recognition network is about 80% compared with the corresponding static image recognition, and the time and power consumption are reduced by 24%, 27.8% and 41% respectively.