Light network construction method for embedded real-time visual target detection

Abstract

The invention discloses a light network construction method for embedded real-time visual target detection. The light network construction method comprises the following steps: step 1) constructing a three-branch output backbone network module Backbone-Tiny based on channel attention mechanism multi-scale feature fusion; step 2), constructing a light pyramid feature fusion network module PAN-Tiny; step 3), constructing a light detection head network module Head-Tiny; and step 4) according to the sequence of the backbone network module Backbone-Tiny, the pyramid feature fusion network module PAN-Tiny and the detection head network module Head-Tiny, sequentially connecting the three local sub-network modules described in the step 1), the step 2) and the step 3) in series to form the whole target detection network. According to the method, the effectiveness of feature extraction of the backbone network module can be greatly improved, and meanwhile, the calculation amount of the feature pyramid fusion network module is remarkably reduced.

Description

technical field [0001] The present invention relates to image target detection and recognition technology, computer vision technology, in particular to a structure lightweight design method for real-time image target positioning and recognition (detection) deep learning model under the environment of embedded mobile terminal equipment. Background technique [0002] The so-called real-time image object detection model in the environment of embedded mobile terminal equipment refers to the development of a volume that can handle image object detection tasks exceeding 40 frames per second by using embedded technology in the environment of mobile equipment terminals without the upper PC. product neural network model. This technology provides important technical support for applications in industry, medicine, national defense, criminal investigation and other fields. [0003] At present, the common method is the target detection model of the YOLO series. Such as YOLOv3, YOLOv4, ...

AI Technical Summary

Problems solved by technology

That is, high precision causes a significant drop in processing speed, and high speed causes a significant drop in detection accuracy.

Images