为了解决电力设备铭牌背景复杂导致图像特征提取难度大的问题,提出了一种基于监督和卷积循环神经网络算法的电力设备铭牌识别方法.用注意力监督和背景监督网络(Attention Supervision Based and Back Ground Suppression Segmentation Network,ASBNet)算法进行文本检测,采用深度残差网络作为网络的骨架,注意力掩膜形成多尺度模块特征和图片细粒度特征,并且用背景抑制模块提高文本前景的感知,提取准确的铭牌图像文本框.将检测到的文本框输入到卷积循环神经网络(Convolutional Recurrent Neural Network,CRNN)中进行文字识别.通过试验结果得知,所提方法与残差网络(Residual Network,RestNet)和YOLOv3 算法模型相比,F值分别提升了 7.56%和 10.38%,说明了所提方法用于电力设备铭牌识别中表现更加优越.
Research on Power Equipment Nameplate Identification Technology Based on ASBNet-CRNN
To solve the problem of difficult image feature extraction caused by the complex background of power equipment nameplate,a method for power equipment nameplate recognition based on supervised and convolutional recurrent neural network algorithm is proposed.Attention supervision based and back ground suppression segmentation network(ASBNet)algorithm is used for text detection,and a deep residual network is used as the backbone of the network.The attention mask forms multi-scale module features and fine-grained image features,and the background suppression module is used to improve the perception of text foreground and extract accurate nameplate image text boxes.The detected text boxes are input into the convolutional recurrent neural network(CRNN)for text recognition.The exper-imental results show that the proposed method outperforms the residual network(RestNet)and YOLOv3 computational models in terms of F-values by 7.56% and 10.38% respectively,indicating that the proposed method performs better in power equipment nameplate recognition.
power equipmentASBNetCRNNnameplate recognitiondeep residual networks
国家科技期刊平台
NSTL
万方数据
