基于图像识别的胶带撕裂检测系统研究

Detection system for belt rip based on image recognition

为了避免出现在煤矿大型带式输送机工作过程中由于带面撕裂引起的漏煤、卡带甚至断带等故障,针对环境的粉尘污染和光照变化等不利影响因素,基于图像识别技术提出了一种具备图像处理、记忆诊断等功能的胶带撕裂检测系统,分析了该系统的组成结构及各部分工作原理。为了便于计算机处理,利用灰度算法将彩色图像转换为灰度图像,为了使特征轮廓更为清晰,利用中值滤波算法锐化特征轮廓,得到边缘更为清晰的图像,利用拉普拉斯算法进一步将图像转换为二值图,提取出胶带表面的特征,最后运用四邻域轮廓跟踪算法获取每一轮廓的边界坐标数组,通过最小外接矩形计算出轮廓的长宽比、面积、周长等参数。为了建立撕裂监测知识库,运用搭建好的胶带撕裂检测系统对20组不同损伤情况的胶带进行分析,将专家对每一图像的每一轮廓外形所给出的健康状况判定与该轮廓的后台计算参数相对应,从而建立系统专家数据库,实现计算机对胶带健康状况的记忆诊断。实验结果表明:系统和算法能够有效计算胶带的破损和撕裂的大小、数量等参数,专家系统能够对胶带当前所处健康状况做出正确的判断;一次滤波仍然在图像上保留部分干扰光点,对后续处理与计算不利,尤其是在轮廓跟踪阶段,这种影响导致的计算缓慢较为突出。

In order to avoid coal leakage,belt sticking and even belt breaking caused by belt surface rip during the operation of large-scale belt conveyor in coal mine,aiming at the adverse factors such as dust pollution and illumination change,a belt rip detection system with image processing,memory diagnosis and other functions is proposed based on image recognition technology.The composition structure and each part of the system are analyzed according to the working principle,in order to facilitate computer processing,the color image is converted into gray image by gray algorithm.In order to make the feature contour clearer,the median filtering algorithm is used to sharpen the feature contour,and the image with clearer edge is obtained.The Laplace algorithm is used to further convert the image into a binary image to extract the features of the belt surface.Finally,the four neighborhood contour tracing algorithm is used to obtain the boundary coordinate array of each contour,and calculates the length-width ratio,area,perimeter and other parameters of the contour through the minimum circumscribed rectangle.In order to establish a tear monitoring expert system,the belt tear detection system is used to analyze 20 groups of belts with different damage conditions,and the judgement on health status of each contour given by the expert on each contour of the images is matched with the particular background computation parameter.The system expert database is established to realize the memory diagnosis of belt health condition by computer.The experimental results show that:the system and algorithm can effectively calculate the size and number of belt breakage and tear,and the expert system can make correct judgment on the current health status of the belt;the primary filtering still retains some interference light points in the image,which is unfavorable to the subsequent processing and calculation,especially in the contour tracking stage,the calculation is sighificantly slowed down.