Improved Angle Steel Character Detection Algorithm Based on DBNet

Aiming at the problems of false detection and missed in manual recording industrial angle steel characters, an improved character algorithm based on DBNet is proposed. First, dataset collected. The uses MobileNetV3 as backbone network, replaces attention mechanism network with a new coordinate to obtain larger regional information enhance ability extract complex features; secondly, weighted bidirectional feature pyramid improves efficiency multi-scale fusion enhances small finally, data augmentation method learning rate optimization strategy are used robustness convergence accuracy model. experimental results show that, compared original algorithm, index increased from 81.42% 99.06% set, which meets needs environment.