一种基于图像识别的燃气表远程直读系统研究

为了解决无线抄表系统中燃气表机械字轮读数与电子计数存在累计误差的问题，提出一种在传统燃气表上加装图像识别抄表模块，利用遗传算法(genetic algorithms，GA)优化的径向基函数神经网络(radical basis function，RBF)进行字轮读数图像识别的实现方法。在图像识别抄表模块中，采用分块迭代算法对图像进行二值化处理，并在远程发送前对图像数据进行压缩，以减少数据传输量；使用最近邻聚类与K均值聚类相结合的算法确定隐层中心位置，为了消除中心宽度对中心值的依赖，采用GA对中心宽度进行优化，引入二次验证机制，进一步减少识别误差；通过通用分组无线业务(general packet radio service，GPRS)方式将识别结果及相应图像发送到管理中心，由管理中心对燃气表读数进一步核对。仿真结果表明，抄表终端功耗较低，抄表正确率超过97%。

Research on remote reading system of gas meter based on image recognition

In order to solve the problem of cumulative error in the mechanical meter reading and electronic counting of the gas meter in the wireless meter reading system, we put forward a method which adds the image processing meter reading module to traditional gas meter. In the module, we can take advantage of radical basis function (RBF) neural network optimized by genetic algorithms (GA) to achieve the purpose of identifying meter reading. In the image processing module, we binarized the image by way of the block iteration algorithm and compressed image data before the remote transmission to reduce the amount of data transmission. The nearest neighbor clustering and K-means clustering algorithm is used to determine the hidden layer center position. In order to eliminate the center-width dependence on the center, the genetic algorithm is used to optimize the center width. Secondary verification mechanism is introduced to further reduce recognition error. In the end, the module sends the recognition results and the corresponding images to the management center by general packet radio service (GPRS). The management center will check the gas meter readings further. The simulation results show that the recognition accuracy is more than 97%.