面向变电站监控界面自动测试的画面识别算法

对变电站监控系统人机界面进行测试验证时，通常采用对比人眼观察到的监控画面与测试指令发送的信息是否一致的方式评估监控软件是否达标，而人眼观察繁杂多变，监控信息的准确率和效率均得不到保证。为了实现变电站监控的自动测试，研究利用图像处理和机器学习技术识别变电站监控画面信息的方法。提出一种基于最佳图元的模板匹配方法解决画面中不同尺寸电气图元的自动定位问题；针对监控画面中拓扑特点提出FHOG算子并提高监控画面和图元状态的识别速度；针对汉字左右体结构分离和告警信息画面中的字符粘连等问题，提出分割识别协同的算法定位字符，并使用深度卷积神经网络进行识别。经线下实验验证了各个单元算法在实际变电站监控图像上的有效性。设计一套测试系统，经线上测试总体图元识别准确率达到96.04%。

Substation Monitoring Picture Recognition Algorithm forAutomatic Human-machine Interface Verification

When testing and verifying the man-machine interface of substation monitoring system, it is common to assess whether the monitoring software is up to standard by comparing the monitoring picture observed by the human eye with the information sent by the test command, but the accuracy and efficiency of the human eye in observing the complex and variable monitoring information is not guaranteed. In this paper, we design a method to automatically identify information on substation monitoring pictures using image processing and machine learning techniques. A template matching method based on the best primitive is proposed to solve the problem of automatic positioning of electrical primitive in the picture.The FHOG operator is proposed to describe the topological features of the picture and speed up the recognition of the monitoring pictures and primitives. For problems such as the separation of the left and right body structure of Chinese characters and the sticking of characters in the warning message picture, an algorithm for segmentation and recognition of synergies is proposed to locate characters and deep convolutional neural networks are used for recognition. The effectiveness of the method is verified in the actual substation monitoring pictures. We also design an online verification system, obtaining the recognition accuracy of 96.04%.