Curved Text Detection Based on Graph Convolution Context Enhancement
- Resource Type
- Conference
- Authors
- Mao, Xinyue; Zhou, Hongyang; Wang, Jiapei; Zhang, Yuan
- Source
- 2022 China Automation Congress (CAC) Automation Congress (CAC), 2022 China. :2097-2102 Nov, 2022
- Subject
- Aerospace
Communication, Networking and Broadcast Technologies
Components, Circuits, Devices and Systems
Computing and Processing
Power, Energy and Industry Applications
Robotics and Control Systems
Signal Processing and Analysis
Transportation
Coils
Convolution
Computational modeling
Production
Feature extraction
Real-time systems
Steel
text detection
industrial scenarios
graph convolution
lightweight networks
parameter count
- Language
- ISSN
- 2688-0938
Text detection in industrial scenarios faces a major challenge: how to design smaller models while maintaining high detection accuracy. We propose a curved text detection model based on graph convolution context enhancement. By using the lightweight backbone MobileNetV2 to reduce the parameter count, and using the graph convolution context module to enhance context modeling ability. The target scene of this paper is the detection of painting characters on steel coils in the production process. Experiments on the Total-Text and the steel coil painting characters dataset show that the proposed method can maintain the optimal detection results with fewer parameters. Graph convolution context module makes the F-measure of this model on the Total-Text test set optimal among the methods based on lightweight networks such as ResNet18 and smaller networks. On the steel coil painting characters dataset, the F-measure of our method is also ahead of PAN and PSENet, and achieves real-time performance.