论文总字数：47324字

摘 要

机器视觉代替人眼进行配电柜检测提高了智能化水平与生产效率。本文研究了配电柜设备开关状态的机器视觉检测，需要完成亮灯形状与颜色检测和电气柜标签光学字符识别两大工作。在电力系统的变电站及类似的工作场景有重要应用价值。

本研究主要分为两个部分，第一部分采用基于OpenCV的图像预处理、图像形态学与特征提取等方法实现对实地小车拍摄的图片中与视频帧中亮灯形状与颜色的检测；第二部分首先采用基于EAST神经网络的算法，检测光学字符所在位置，接着对比采用基于LSTM神经网络的Tesseract识别引擎，识别并输出字符。整体环境搭建在OS X下，编程语言选用C ，运用OpenCV与Tesseract两大现有平台实现该两大工作

仿真实验结果表明，传统图像识别与处理的方法取得了准确率不低于95%的良好效果，可直接运用于实际的实时检测；基于EAST神经网络检测展现了准确率不低于90%的检测效果，而基于LSTM神经网络的Tesseract运用ROI和正则化匹配对其改进，使之展现了准确率不低于85%的识别效果，亦可直接运用于实际的实时识别。

关键词：颜色检测 形状检测 EAST LSTM Tesseract

Machine Vision Inspection of Switchboard Equipment Switching Status

Abstract

Machine vision instead of the human eye for power distribution cabinet detection improves the level of intelligence and production efficiency. In this paper, the machine vision inspection of the switch state of the power distribution cabinet equipment is studied, and the two tasks of lighting shape and color inspection and electrical cabinet label optical character recognition(OCR) are accomplished. It has important application value in substation of power system and similar work scenarios.

This research is mainly divided into two parts. The first part adopts OpenCV-based image preprocessing, mathematical morphological image processing and feature extraction to realize the inspection of the shape and color of the lights in the static pictures and the dynamic videos taken by the field vehicle; the second part adopts the algorithm based on EAST neural network to detect the position of optical characters, and then the Tesseract recognition engine based on LSTM neural network is used to identify and output characters. The overall environment is built under OS X, the programming language is C , and the two existing platforms, OpenCV and Tesseract, are used to achieve the two major tasks.

The simulation results show that the traditional image recognition and processing method has achieved satisfying results with an accuracy above 95%, which can be directly applied to industrious real-time inspection; the EAST neural network based detection demonstrates an accuracy above 90%, while the LSTM neural network-based Tesseract, adopting ROI and regular expression to improve it, displays the recognition rate with an accuracy above 85%, and can also be directly applied to practical real-time recognition.

Key Words: Color Detection Shape Detection EAST LSTM Tesseract

目 录

摘 要 I

Abstract II

第一章 绪论 1

1.1研究背景及意义 1

1.2 国内外研究现状及发展趋势 1

1.2.1人工目视检测法 2

1.2.2辅助开关检测法 2

1.2.3激光感应检测法 2

1.2.4视觉检测法 2

1.3研究问题与研究手段 4

1.3.1研究问题 4

1.3.2研究手段 4

1.4 章节安排 4

第二章 颜色与形状及字符识别的技术基础 6

2.1机器视觉 6

2.2视频图像预处理 6

2.3机器视觉中的颜色识别 7

2.4机器视觉中的形状识别 9

2.5人工神经网络原理与相关种类 9

2.5.1卷积神经网络 11

2.5.2 长短时记忆网络 11

2.6人工神经网络在光学字符检测识别与图像分析上的应用 12

2.6.1基于人工神经网络的图像处理 12

2.6.2光学字符识别 13

第三章 基于颜色与形状识别的配电柜设备开关状态检测 14

3.1 基本原理 14

3.2 配电柜开关状态检测总体设计 14

3.3 CMake对OpenCV的设置 16

3.4开关柜状态检测程序设计 17

3.5实验结果 19

第四章 配电柜设备字符定位与识别 24

4.1基本原理 24

4.1.1 Efficient and Accuracy Scene Text detection(EAST)检测原理 24

4.1.2 LSTM识别原理 25

4.2 CMake对Tesseract的设置 26

4.3算法及软件的选用 27

4.3.1 EAST 27

4.3.2 Tesseract 27

4.4光学字符检测程序设计 28

4.5光学字符识别程序设计 29

4.5.1改进前 29

4.5.2改进后 30

4.6实验结果 31

4.6.1 EAST光学字符检测 31

4.6.2 Tesseract光学字符识别 32

第五章 论文总结及展望 35

5.1 论文总结 35

5.2 研究展望 35

5.2.1电气柜开关状态检测 35

5.2.2光学字符识别程序设计 35

参考文献 36

附录 39

致谢 55

第一章 绪论

1.1研究背景及意义

视觉检测在工业生产领域有着越来越广泛的应用。尽管很有挑战性，机器视觉检测甚至尝试没有标签或部分标签图像[1]。现代机器视觉领域，损失函数最优化问题是热点之一[2]。

请支付后下载全文，论文总字数：47324字