FPGA Implementation Method of Kernel Function Extreme Learning Machine Classifier

Abstract

The invention discloses an FPGA implementation method of a kernel function extreme learning machine classifier. The FPGA implementation method comprises the following steps: firstly, preprocessing an original classified sample on a PC to obtain a sample, then transmitting the sample into the FPGA through an RS232 port by the PC, storing the sample into an RAM by the FPGA, and determining a decision function and a topological structure of the learning machine according to the characteristic number and the sample number of a training sample. The kernel function extreme learning machine has good classification capability, simple operation, high training speed and good generalization, and also can avoid the risk of falling into a local minimum. The innovation of this invention is the use of parallel and serial combined programming, which can effectively reduce the use of resources; the FPGA for inversion of partitioned matrix of dimension reduction method is implemented; the FPGA implementation method is suitable for the inversion of matrixes in arbitrary dimensions, is easy and convenient in modification, can effectively improve the work efficiency, can use binary numbers with different bit widths according to the precision requirements, and can effectively reduce the resource consumption while maintaining the accuracy.

Description

Technical field: [0001] The invention belongs to the technical field of pattern recognition, and in particular relates to an FPGA implementation method of a kernel function extreme learning machine classifier. Background technique: [0002] An artificial neural network is a neural network that is artificially constructed on the basis of human understanding of its brain neural network and can achieve certain functions. It is actually a complex network composed of a large number of simple neurons connected to each other, with highly nonlinear characteristics, parallelism, and a system capable of complex logic operations and nonlinear relationships. However, the training of the feedforward neural network mainly adopts the gradient descent algorithm, and all weights need to be adjusted, which limits the training speed of the neural network. G.B.Huang conducted research on this and proposed an extreme learning machine (ELM) algorithm, which randomly assigns the input weights and...

AI Technical Summary

Problems solved by technology

[0003] At present, the kernel function extreme learning machine is only programmed on the computer. Although the computer is highly flexible, it is not suitable for the neural network due to the serial operation of the computer.

Images