Open AccessResearch on Image Recognition of Power Inspection Robot Based on Improved YOLOv3 Model
Author(s) -
Wei Xiong,
Sha Yang,
Zhao Zhang,
Liang Chen,
Shuxin Huang
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1486/4/042034
Subject(s) - computer science , artificial intelligence , convolution (computer science) , robot , feature (linguistics) , image (mathematics) , scale (ratio) , feature extraction , power (physics) , pattern recognition (psychology) , deep learning , computer vision , artificial neural network , philosophy , linguistics , physics , quantum mechanics
YOLO series models are widely used in power inspections, but they are prone to miss inspections for small targets and diverse targets. In response to this defect, an improved network model of YOLOv3-g suitable for GPU cores and an improved network model of YOLOv3 mini suitable for CPU cores are proposed. By reducing the number of small and medium-scale targets, the number of convolution kernels is increased, and the small and medium-scale targets are increased. The intensity of feature extraction reduces the amount of network calculations and better solves the problem of inaccurate detection and recognition of small targets in electric robot scenes.