Abstract:Aiming at the problem of high-precision recognition of signboard in power patrol,a model of power signboard detection and recognition based on improved PPYOLOE is proposed.Firstly,the RepResBlock module is modified to enhance the feature expression ability of single parameter convolution kernel,then the weight dimension of the model is increased,and the generalization ability of the whole network is raised.Secondly,the CIoU Loss function is introduced,which solves the problem of slow convergence when the prediction box does not intersect the real box,and ensures the aspect ratio of prediction box and the real box closer,which upgrades the regression accuracy of the prediction box.Thirdly,the data enhancement Mosaic method is optimized to reduce the error detection rate of negative samples and reinforce the accuracy and robustness of the model.The experimental results show that the proposed method significantly improves the detection model performance,with an mAP_0.5 of 98.4%.Moreover,the model volume is compressed to 26.1% after quantification and distillation,and self-made sample library enables both the text detection and recognition accuracy to exceed 90%.