【作者】 张伟；

【导师】 杨公平；

【作者基本信息】 山东大学 ， 软件工程（专业学位）， 2022， 硕士

【摘要】 传统机械水表因其低廉的价格、精准的计量、成熟的技术以及稳定的供应和销售渠道,使其在市场中仍处于主导地位。虽然智能水表正在逐步兴起,但替换传统机械水表必须投入大量的人力、物力和财力,购买新设备,搭建对应的水务系统,其改造过程必定会很缓慢。而随着智能时代的到来,人们对智能化的需求也越来越高,但自来水公司仍采用传统的人工抄表方式,效率低下,环节繁琐,还会因人工不可避免的误操作使用户不满,从而影响企业形象。在这种情况下,设计一种自动抄表识别系统便显得尤为重要,不仅能降低企业的人工成本,工作效率和准确性也会大大提升。水表数字识别过程可分为三个阶段:数字区域定位,字符分割和字符识别。在数字区域定位阶段,以往研究多建立在严格的先决条件下,限制了水表的样式,对光照条件、拍摄环境、拍摄距离等均有严格的要求。然而在实际的生活使用中,由于水表的生产批次、所处位置、拍照时间、表盘污染情况各不相同,忽略以上因素会对实际的生产使用产生很大影响,开发出的系统也不会具有较好的鲁棒性。为了能在以上各种不确定因素下,仍能精确定位水表数字区域,受文本检测思想的启发,本文提出了一种基于文本检测算法DBNet的网络模型,同时使用数据增强技术,实现了数字区域定位。实验结果验证了此算法的有效性,对于不同样式的水表以及在各种复杂环境因素的条件下均取得了很好的定位效果。在数字识别过程中,字符分割的好坏程度直接影响着识别结果的准确与否。在字符分割阶段,针对数字区域光照不均的问题,本文通过实验对比,最终选择使用局部自适应阈值算法生成二值图像,然后提出了一种基于形态学、连通区域与映射的字符分割方法,实现了数字区域字符分割,实验结果表明对于存在光照不均、表盘污染的数字区域,该方法均取得了很好的分割效果。在最后的字符识别阶段,本文使用模板匹配法用于数字识别。针对因水表样式不同,生产批次不同造成字符间隔存在细微差异以及使用同一个固定模板导致识别不正确的情况,本文基于以往研究进一步改进了匹配模板,改进后的模板同时也解决了以往研究中将整字符和半字符分开匹配的问题,只需要将半字符拼接后形成新的待匹配字符即可使用此模板进行匹配。最后为了解决半字符拼接形成的新字符与整字符尺寸不一致导致匹配精度下降的问题,本文提出了一种多尺度模板匹配方法,经实验验证表明,该方法可以有效地识别数字,不但效率高而且准确率很高。更多还原

【Abstract】 Traditional mechanical water meters are still dominant in the market due to their low prices,accurate metering,mature technology,and stable supply and sales channels.Although smart water meters are gradually emerging,replacing traditional mechanical water meters must invest a lot of manpower,material and financial resources,purchase new equipment,and build corresponding water systems.The transformation process will certainly be very slow.With the advent of the era of intelligence,people’s demand for intelligence is getting higher and higher,but water companies still use the traditional manual meter reading method,which is inefficient and cumbersome,and users will be dissatisfied due to inevitable manual misoperation,thus affecting the corporate image.In this case,it is particularly important to design an automatic meter reading identification system,which can not only reduce the labor cost of the enterprise,but also greatly improve the work efficiency and accuracy.The process of water meter digital recognition can be divided into three stages:digital area positioning,character segmentation and character recognition.In the stage of digital area positioning,previous studies were mostly based on strict prerequisites,which limited the style of water meters,and had strict requirements on lighting conditions,shooting environment,shooting distance,etc.However,in actual life use,due to the different production batches,locations,photographing times,and dial contamination of water meters,ignoring the above factors will have a great impact on actual production and use,and the developed system will not Has better robustness.In order to be able to accurately locate the digital area of the water meter under the above various uncertainties,inspired by the idea of text detection,this paper proposes a network model based on the text detection algorithm DBNet,and uses data enhancement technology to achieve digital area localization.The experimental results verify the effectiveness of the algorithm,and it has achieved good positioning results for different styles of water meters and under various complex environmental factors.In the process of digital recognition,the quality of character segmentation directly affects the accuracy of the recognition results.In the character segmentation stage,in view of the problem of uneven illumination in the digital area,this paper finally chooses to use the local binarization algorithm to obtain binary images through experimental comparison,and then proposes a character segmentation method based on morphology,connected areas and mapping.Character segmentation in digital areas.The experimental results show that this method has achieved good segmentation results for digital areas with uneven illumination and dial pollution.In the final character recognition stage,this paper uses the template matching method for digit recognition.Aiming at the slight difference in character interval caused by different water meter styles,different production batches,and incorrect recognition caused by the use of the same fixed template,this paper further improves the matching template based on previous research.The improved template also solves the problem in previous research.For the problem of matching whole characters and half-characters separately,you only need to splicing the half-characters to form a new to-be-matched character to use this template for matching.Finally,in order to solve the problem that the size of the new characters formed by half-character splicing is inconsistent with the whole character,which leads to the decrease of matching accuracy,a multi-scale template matching method is proposed in this paper.The experimental verification shows that this method can effectively recognize numbers,not only with high efficiency but also with high accuracy.更多还原