基于CNN的红外图像边缘检测算法的FPGA实现

提出了一种以现场可编程门阵列为硬件处理器实现基于细胞神经网络的红外图像边缘检测方法.首先利用simulink的算法行为特性搭建红外图像输入模块,获得相关的红外图像头信息并对红外图像像素值范围进行相应变化,然后根据细胞神经网络模板所创建的查找表设计单个细胞元软核,再利用细胞神经网络阵列的规则性和互联的局域性,将单个细胞元软核扩展成细胞神经网络阵列.最后采用modelsim将细胞神经网络阵列与红外图像输入、输出模块相关联,从而达到实时处理的效果.实验结果表明:基于现场可编程门阵列为硬件处理器平台实现的细胞神经网络对红外图像进行边缘检测取得了较好的效果,且与MATLAB软件仿真的结果进行对比得出两者只有极其微小的差别.在Xilinx公司Virtex-6系列的现场可编程门阵列平台上,综合后占用极少资源的情况下得到142.693 MHz的最高频率,并且达到了2.378 Mpixels/sec处理速度.

The Implementation of Infrared Image Edge Detection Algorithm Based on CNN on FPGA

A novel infrared image edge detection algorithm based on cellular neural networks on FPGA is proposed. First, the infrared image input module is to build with simulink. The relevant information of infrared image head is acquired, and the infrared image pixel value range is adjusted. Then CNN IP core is designed by lookup table which is created by the template of cellular neural networks. With the regularity and interconnection locality of cellular neural network array, the CNN IP core will be expanded into the cellular neural network array. Then the cellular neural network array is related with the infrared image input and output module by modelsim, so that the infrared image will be processed in real time. The experimental results showed: In the field programmable gate array hardware processor platforms such as the Virtex-6 FPGA of the Xilinx, the infrared image edge detection algorithm will be implementationed with cellular neural networks. The highest frequency of 142.693MHz is got, and the system processing speed of 2.378Mpixels/sec is reached.