基于可变形卷积神经网络的数字仪表识别方法  

Digital Instrument Identification Method Based on Deformable Convolutional Neural Network

作　　者：郭兰英[1] 韩睿之 程鑫[1] GUO Lan-ying;HAN Rui-zhi;CHENG Xin(School of Information Engineering,Chang’an University,Xi’an 710064,China)

机构地区：[1]长安大学信息工程学院,西安710064

出　　处：《计算机科学》2020年第10期187-193,共7页Computer Science

基　　金：陕西省重点研发计划(2019NY-163);陕西省交通科技项目(14-23K);中央高校基本科研业务费专项资金项目(300102329101,310824175004)。

摘　　要：目前,对于数显仪表的识别,多采用传统的图像处理及机器学习等方法,在复杂多变的应用场景中,其对字符、数字的识别准确率低,难以满足实时应用的要求。针对以上问题,将传统图像处理技术与深度学习方法相结合,提出了一种基于可变形卷积神经网络的数显仪表示数分割与识别方法。该方法包含图像预处理、字符分割与识别等步骤。首先,使用GrayWorld算法对待识别图像进行亮度均衡,并通过彩色分割提取屏幕区域;其次,对图像进行形态学操作,以便使用投影直方图法完成字符与对应小数点的整体分割;最后,设计并训练了一种可变形卷积神经网络对字符进行识别,优化了卷积神经网络感受野几何结构固定的内在问题。实验结果表明,加入可变形卷积有效提高了图像的识别准确率和网络的收敛速度;该方法的整体识别准确率达到99.45%,检测速度为10FPS,能够满足实际应用需求。At present,traditional image processing methods and machine learning methods are adopted for the identification of digital display instruments,which have disadvantages such as low recognition accuracy for both characters and numbers in complicated scenarios,and difficulty to meet real-time application requirements.Aiming at the problems above,combining traditional image processing technology and deep learning methods,a method of segmentation and recognition of digital display instrument based on deformable convolutional neural network is proposed.This method includes steps such as image preprocessing,character segmentation and image recognition.Firstly,the GrayWorld algorithm is applied to perform brightness equalization on the image to be recognized for the further using of color segmentation to extract the screen area.Secondly,the projected histogram method is implemented to realize the unified segmentation of characters with its corresponding decimal point after performing morphological operation on the image.Finally,a deformable convolutional neural network is proposed and trained for character recognition,which optimizes the endogenous geometry restriction of receptive field in convolutional neural networks.The experimental results indicate that the addition of deformable convolution effectively improves the accuracy of image recognition and the convergence speed of the network,and the accuracy of the overall recognition method reaches 99.45%and the detection speed is 10FPS,which can meet the requirements of practical applications.

关 键 词：图像处理 字符识别 可变形卷积神经网络 投影直方图 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]