LAUSR

Target recognition system based on machine learning has the problems of long delay, high power-consuming and high cost, which cause it difficult to be promoted in some small embedded devices. In order to develop a target recognition system based on machine learning that can be utilized in small embedded device, this paper analyzes the commonly used design process of target recognition, the training process of machine learning algorithms, and the working method of FPGA to accelerate the algorithm. In the end, it offers a new solution of target recognition system based on machine learning hardware accelerator. In the solution, the training process of target recognition algorithm based on machine learning is completed in GPU, and then the algorithm is porting to the logic part of SOC in the form of hardware accelerator. The solution be widely used in different needs of the target recognition scenario with the advantage of effectively reduce the system delay, power consumption, size.