Image processing method based on fpga-accelerated convolutional neural network framework

Abstract

The invention discloses a picture processing method based on an FPGA accelerated convolutional neural network framework, which mainly solves the problems of low resource utilization and slow speed in the prior art. The scheme is: 1) Calculate the fixed value of image segmentation according to the designed image parameters and FPGA resource parameters; 2) Determine the number of DDR3 according to the fixed value of the image, and allocate block ram resources; 3) Construct the volume according to 1) and 2) A product neural network framework, which includes a picture storage module, a picture data distribution module, a convolution module, a pooling module, a picture storage back DDR3 module, and an instruction register group; 4) Each module obtains control instructions from the instruction register group through handshake signals , and cooperate with each other to process the picture data according to the control instructions. The invention improves the resource utilization rate and acceleration effect through the FPGA-accelerated convolutional neural network framework, and can be used for image classification, target recognition, speech recognition and natural language processing.

Description

technical field [0001] The invention belongs to the technical field of computer design, in particular to a convolutional neural network implementation method, which can be used for image classification, target recognition, speech recognition and natural language processing. Background technique [0002] With the advancement of integrated circuit design and manufacturing technology, the field programmable gate array with high-speed, high-density programmable logic resources has been developed rapidly, and the integration level of a single chip is getting higher and higher. In order to further improve FPGA performance, mainstream chip manufacturers have integrated digital signal processing chips DSP custom computing units with high-speed digital signal processing capabilities inside the chips. It is widely used in image processing, network communication and information security, bioinformatics and other application fields. [0003] Convolutional neural network CNN is a struct...

AI Technical Summary

Problems solved by technology

[0005] In the current patent and research direction, the OpenCL programming language is basically used as the core of the construction. The purpose is to reduce the implementation time of converting the convolutional neural network algorithm into a hardware description language, but it does not involve the acceleration of the hardware description language code in the FPGA algorithm. At the same time, since the OpenCL programming language is not the language that actually runs on the FPGA, the running speed of the actual FPGA is not ideal.

At present, the existing technology based on OpenCL programming mainly focuses on the acceleration of the DSP module in the FPGA, and has not implemented the convolutional neural network algorithm as a whole and optimized the underlying hardware description language, and cannot make full use of the computing resources of the FPGA. It leads to an increase in FPGA calculation time, and the acceleration effect is not obvious

Images