Hardware architecture of deep learning-based target detection algorithm and execution method thereof

Abstract

The invention provides hardware architecture of a deep learning-based target detection algorithm and an execution method thereof. The hardware architecture comprises an input cache, a line cache, a register matrix, a convolution calculation kernel, an output cache and a full-connection calculation kernel, which are arranged on an FPGA, wherein the input cache is used for caching data of an input layer of the deep learning-based target detection algorithm; the line cache comprises k storage units and is used for caching output data of k lines of input caches; the register matrix comprises k*k registers; the convolution calculation kernel is used for carrying out convolution calculation according to data, of the k*k registers, output by the register matrix in each clock period; the output cache is used for storing an output result of the convolution calculation kernel; and the full-connection calculation kernel is used for carrying out calculation to obtain a final result of target detection. According to the hardware architecture of the deep learning-based target detection algorithm and the execution method thereof, real-time calculation of the deep learning-based target detection algorithm can be realized under FPGA hardware architecture.

Description

technical field [0001] The present invention relates to the technical field of FPGA, in particular to a hardware architecture of an object detection algorithm based on deep learning and an execution method thereof. Background technique [0002] Field Programmable Gate Array (Field-Programmable Gate Array, FPGA) is an integrated circuit chip that can be programmed after production. The circuit in the FPGA chip provides programmable nodes, which can redefine the circuit logic according to user settings. Compared with traditional processing chip CPUs, FPGAs can provide highly optimized circuits for specific problems, improving computing performance by hundreds of times. Compared with traditional integrated circuit chip ASIC, FPGA can provide more flexible computing solutions. [0003] Target detection, also called target extraction, is an image segmentation based on target geometric and statistical features, which combines target segmentation and recognition into one. Object...

AI Technical Summary

Problems solved by technology

In robots, drones, satellites and other mobile devices that urgently need artificial intelligence support, due to limited computing resources and space, it is impossible to support the realization of such deep learning-based target detection algorithms.

Images