MobileNet-SSD target detection device and method based on FPGA acceleration

Abstract

The invention provides a MobileNet-SSD target detection device and method based on FPGA acceleration, and the method comprises the following steps: Step 1: obtaining an initial image of a to-be-detected target, and storing the initial image data of the target and a convolutional neural network weight file; 2, adding a detection layer into the Mobilenet-SSD convolutional neural network, sharing an accelerator among different network layers, and arranging an on-chip buffer area on the accelerator; 3, operating the Mobilenet-SSD convolutional neural network through a parallel block parameter optimization method, and outputting block coefficients of a feature map channel and a convolution kernel channel of the target image; 4, through a pipeline algorithm, completing target feature extraction, bounding box regression and object category prediction of the Mobilenet-SSD convolutional neural network; 5, performing post-processing on target feature extraction, bounding box regression and object category prediction information obtained by the convolutional neural network; and 6, testing a target detection result on the target image to be detected. According to the invention, the improved Mobilenet-SSD is adopted as a target detection algorithm, and high-precision target detection is realized.

Description

technical field [0001] The invention relates to the technical field of target detection and recognition, in particular to an FPGA-accelerated MobileNet-SSD target detection device and method. Background technique [0002] Target detection is widely used in civil and military fields such as artificial intelligence, medical research, and national defense security. The target detection algorithm based on deep learning uses convolutional neural network CNN (Convolutional Neural Network) to extract features and complete image classification and positioning. Compared with the traditional algorithm, the speed and speed are greatly improved, but the convolutional neural network often has a huge amount of parameters and calculations, and the parameters and structure of the network layer are changeable, which makes it difficult for the target detection algorithm to be applied in limited resources and high requirements. Embedded applications with processing speed and low power consumpt...

AI Technical Summary

Problems solved by technology

[0003] There are two main difficulties in deploying the most advanced CNN-based target detection algorithm to FPGA to achieve hardware acceleration: (1) The convolutional neural network-based target detection algorithm usually has a complex network model and a large number of parameters. And the amount of calculation is large, but the FPGA storage resources, computing resources and system bandwidth are extremely limited, which causes problems in the real-time performance of the system

Images