输送带纵向撕裂一体化双目视觉检测方法研究
发布时间：2018-07-29 16:34
【摘要】：输送机是现代采矿工业的重要运输工具。在煤炭运输中,煤矸石、金属材料等坚硬物质经常混合在煤炭中,可能会使得输送带产生纵向撕裂,这样的猝发事故通常会导致采矿设备和生产的停工,造成巨大的经济损失,因此我们需要实时、可靠的检测输送带纵向撕裂。近年来,由于机器视觉可以提高检测效率和精度,在输送带故障检测中,视觉检测成为一个重要的研究方向。本文基于机器视觉,综合红外成像技术和可见光成像技术的互补特性,提出了一种基于红外与可见光融合的输送带纵向撕裂一体化双目视觉检测方法,并依据输送带纵向撕裂视觉检测方法现有问题,提出本文检测方法的设计方案,主要包括三部分:输送带图像采集,输送带图像预处理,输送带图像纵向撕裂特征提取与识别。在输送带图像采集中,基于红外与可见光融合技术,论文提出一种新的一体化双目视觉传感装置采集红外与可见光融合图像,该传感装置通过棱镜分光的方法将从同一镜头入射的同轴光分为红外光和可见光,并分别进入两个感光芯片,可以同时捕获同一场景下的红外信息和可见光信息,在图像融合前无需配准处理。本文研究了该传感装置的成像原理,并在MATLAB上对其成像过程进行理论仿真研究,验证其可行性和有效性。由于论文只对提出的一体化双目视觉传感装置进行了理论仿真研究,因此在输送带图像采集实验部分,首先采用红外相机和可见光相机分别采集红外图像和可见光图像,然后再进行红外与可见光图像的配准融合,最终实现输送带红外与可见光融合图像的采集。在实验室搭建图像采集实验平台,实现输送带在撕裂状态、正常状态和划痕状态的融合图像的采集。输送带融合图像采集完成后,为使得检测效果更可靠,需对其进行预处理操作。论文分析了输送带纵向撕裂的特征,并据此对图像进行了一系列的预处理,包括去除图像噪声,增强图像对比度,突出检测中感兴趣的区域,提取出撕裂目标信息。预处理后的图像视觉效果较好,撕裂目标较为突出,为后续纵向撕裂特征提取与识别做了较好准备。在得到的预处理图像的基础上,本文采用投影法提取出图像的投影特征,并据此计算出图像的几何特征,即输送带纵向撕裂参数:撕裂长度、宽度和面积。根据各类型输送带图像投影特征的特点和纵向撕裂参数设定识别阈值并规定纵向撕裂的识别规则,最终将输送带图像归为:撕裂状态、正常状态、划痕状态,实现不同状态的输送带图像的识别检测。在MATLAB编程软件平台上对采集得到的输送带图像进行预处理和撕裂特征提取与识别处理。实验结果表明本文提出的输送带纵向撕裂一体化双目视觉检测方法能够识别撕裂与划痕,能够预测潜在的撕裂,检测精确度为96%以上,单帧图像检测时间小于21ms,是一种可靠的、实时的在线检测方法。
[Abstract]:Conveyor is an important transportation tool in modern mining industry. In coal transportation, coal gangue, metal materials and other hard materials are often mixed in coal, which may lead to longitudinal tearing of conveyor belt. Such sudden accidents usually lead to the stoppage of mining equipment and production, resulting in huge economic losses. Therefore, we need real-time, reliable detection of belt longitudinal tear. In recent years, because machine vision can improve detection efficiency and precision, vision detection has become an important research direction in conveyor belt fault detection. Based on the complementary characteristics of machine vision, infrared imaging technology and visible light imaging technology, a binocular vision detection method for longitudinal tear of conveyor belt based on infrared and visible light fusion is proposed. According to the existing problems of the visual detection method of belt longitudinal tear, the design scheme of this method is put forward, which includes three parts: belt image collection, belt image preprocessing, belt image longitudinal tear feature extraction and recognition. Based on infrared and visible light fusion technology, a new integrated binocular vision sensor is proposed to collect infrared and visible light fusion images. The sensing device divides the incident coaxial light from the same lens into infrared light and visible light by means of prism, and enters into two photosensitive chips respectively, which can simultaneously capture infrared and visible light information in the same scene. There is no need for registration before image fusion. In this paper, the imaging principle of the sensor is studied, and its imaging process is simulated on MATLAB to verify its feasibility and effectiveness. Due to the theoretical simulation of the integrated binocular vision sensor proposed in this paper, the infrared image and the visible light image are collected by infrared camera and visible light camera respectively in the experiment part of the conveyor belt image acquisition. Then the infrared and visible image registration fusion, finally achieve the conveyor belt infrared and visible light fusion image acquisition. An image acquisition experiment platform is set up in the laboratory to realize the fusion image acquisition of conveyor belt in tearing state, normal state and scratch state. In order to make the detection more reliable, the belt fusion image acquisition needs to be preprocessed. In this paper, the characteristics of longitudinal tear of conveyor belt are analyzed, and a series of preprocessing is carried out on the image, including removing image noise, enhancing image contrast, highlighting the region of interest in detection, and extracting the information of tearing target. After preprocessing, the visual effect of the image is better, and the tear target is more prominent, which makes a good preparation for the subsequent longitudinal tear feature extraction and recognition. On the basis of the preprocessed images, the projection features of the images are extracted by projection method, and the geometric features of the images are calculated, that is, the longitudinal tearing parameters of the conveyor belt: tear length, width and area. According to the characteristics of projection features of each type of conveyor belt image and the parameters of longitudinal tear, the identification threshold is set and the recognition rule of longitudinal tear is stipulated. Finally, the conveyor belt image is classified as: tearing state, normal state, scratch state, and so on. Realization of different states of the conveyor belt image recognition and detection. The preprocessing, tearing feature extraction and recognition processing of the collected conveyor belt images are carried out on the platform of MATLAB programming software. The experimental results show that the proposed binocular visual detection method for longitudinal tear of conveyor belt can recognize tear and scratch, and can predict potential tear. The detection accuracy is over 96%, and the detection time of single frame image is less than 21ms. it is a reliable method. Real-time online detection method.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TD528.1;TP391.41
本文编号：2153304
[Abstract]:Conveyor is an important transportation tool in modern mining industry. In coal transportation, coal gangue, metal materials and other hard materials are often mixed in coal, which may lead to longitudinal tearing of conveyor belt. Such sudden accidents usually lead to the stoppage of mining equipment and production, resulting in huge economic losses. Therefore, we need real-time, reliable detection of belt longitudinal tear. In recent years, because machine vision can improve detection efficiency and precision, vision detection has become an important research direction in conveyor belt fault detection. Based on the complementary characteristics of machine vision, infrared imaging technology and visible light imaging technology, a binocular vision detection method for longitudinal tear of conveyor belt based on infrared and visible light fusion is proposed. According to the existing problems of the visual detection method of belt longitudinal tear, the design scheme of this method is put forward, which includes three parts: belt image collection, belt image preprocessing, belt image longitudinal tear feature extraction and recognition. Based on infrared and visible light fusion technology, a new integrated binocular vision sensor is proposed to collect infrared and visible light fusion images. The sensing device divides the incident coaxial light from the same lens into infrared light and visible light by means of prism, and enters into two photosensitive chips respectively, which can simultaneously capture infrared and visible light information in the same scene. There is no need for registration before image fusion. In this paper, the imaging principle of the sensor is studied, and its imaging process is simulated on MATLAB to verify its feasibility and effectiveness. Due to the theoretical simulation of the integrated binocular vision sensor proposed in this paper, the infrared image and the visible light image are collected by infrared camera and visible light camera respectively in the experiment part of the conveyor belt image acquisition. Then the infrared and visible image registration fusion, finally achieve the conveyor belt infrared and visible light fusion image acquisition. An image acquisition experiment platform is set up in the laboratory to realize the fusion image acquisition of conveyor belt in tearing state, normal state and scratch state. In order to make the detection more reliable, the belt fusion image acquisition needs to be preprocessed. In this paper, the characteristics of longitudinal tear of conveyor belt are analyzed, and a series of preprocessing is carried out on the image, including removing image noise, enhancing image contrast, highlighting the region of interest in detection, and extracting the information of tearing target. After preprocessing, the visual effect of the image is better, and the tear target is more prominent, which makes a good preparation for the subsequent longitudinal tear feature extraction and recognition. On the basis of the preprocessed images, the projection features of the images are extracted by projection method, and the geometric features of the images are calculated, that is, the longitudinal tearing parameters of the conveyor belt: tear length, width and area. According to the characteristics of projection features of each type of conveyor belt image and the parameters of longitudinal tear, the identification threshold is set and the recognition rule of longitudinal tear is stipulated. Finally, the conveyor belt image is classified as: tearing state, normal state, scratch state, and so on. Realization of different states of the conveyor belt image recognition and detection. The preprocessing, tearing feature extraction and recognition processing of the collected conveyor belt images are carried out on the platform of MATLAB programming software. The experimental results show that the proposed binocular visual detection method for longitudinal tear of conveyor belt can recognize tear and scratch, and can predict potential tear. The detection accuracy is over 96%, and the detection time of single frame image is less than 21ms. it is a reliable method. Real-time online detection method.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TD528.1;TP391.41
【参考文献】
相关期刊论文 前10条
1 权亚楠;卜丽静;武文波;;改进的阈值加权平均HSV与小波变换图像融合[J];辽宁工程技术大学学报(自然科学版);2016年01期
2 王国华;刘琼;庄家俊;;基于局部特征的车载红外行人检测方法研究[J];电子学报;2015年07期
3 李锋;阚建霞;;基于Sobel算子的图像快速二维最大熵阈值分割算法[J];计算机科学;2015年S1期
4 周渝人;耿爱辉;张强;陈娟;董宇星;;基于压缩感知的红外与可见光图像融合[J];光学精密工程;2015年03期
5 周靖鸿;周璀;朱建军;樊东昊;;基于非下采样轮廓波变换遥感影像超分辨重建方法[J];光学学报;2015年01期
6 王建勋;;煤矿输送带传输故障实时监测技术[J];工矿自动化;2015年01期
7 周雨薇;杨平吕;陈强;孙权森;;基于MTF和变分的全色与多光谱图像融合模型[J];自动化学报;2015年02期
8 李旭阳;易红伟;齐浩程;;多光谱遥感相机光学系统设计[J];光子学报;2015年03期
9 张鹏;张志辉;;基于分段直方图变换的图像非线性增强[J];光子学报;2014年S1期
10 何海明;齐冬莲;张国月;张建良;;快速高效去除图像椒盐噪声的均值滤波算法[J];激光与红外;2014年04期
相关博士学位论文 前2条
1 卜凡;光学遥感系统的建模仿真及图像处理技术研究[D];中国科学院研究生院（西安光学精密机械研究所）;2014年
2 李寒;基于机器视觉的目标检测在精细农业中的关键技术研究[D];中国农业大学;2014年
相关硕士学位论文 前10条
1 阳婷;基于视频监控的火灾探测系统的研究与实现[D];东华大学;2016年
2 彭宝;基于机器视觉辅助驾驶系统中行人实时检测跟踪研究[D];东华大学;2016年
3 佟卓远;基于机器视觉的前方车辆检测与测距系统设计[D];哈尔滨工业大学;2015年
4 朱妍妍;基于机器视觉的胶管表面缺陷检测系统研究[D];北京理工大学;2015年
5 崔东顺;可见光航拍图像水上桥梁检测算法研究[D];北京理工大学;2015年
6 何伟;基于小波变换和假彩色的医学图像融合[D];北京理工大学;2015年
7 葛世国;基于数学形态学的遥感图像分割算法研究[D];成都理工大学;2014年
8 狄帅;基于FPGA的输送带表面超声检测系统开发[D];华东理工大学;2014年
9 陈永亮;灰度图像的直方图均衡化处理研究[D];安徽大学;2014年
10 韩博;手持式红外与可见光图像融合系统研究[D];南京理工大学;2014年
，本文编号：2153304
本文链接：https://www.wllwen.com/shoufeilunwen/boshibiyelunwen/2153304.html
