基于显著性和深度卷积神经网络的输送带表面故障定位  

Surface fault location of conveyor belt based on saliency and deep convolution neural network

作　　者：赵燕飞[1] 杨彦利[1] 王丽娟[1] 

机构地区：[1]天津工业大学天津市光电检测技术与系统重点实验室,天津300387

出　　处：《工矿自动化》2016年第12期72-77,共6页Journal Of Mine Automation

基　　金：国家自然科学基金资助项目(51274150;51504164);天津市科技支撑计划重点项目(13ZCZDGX01000)

摘　　要：提出了一种基于显著性和深度卷积神经网络的输送带表面故障定位方法。该方法在输送带上、下表面的边缘烙上数字,利用图像处理技术检测输送带图像中的数字来间接定位输送带表面故障。首先,将采集的输送带图像进行高斯滤波、灰度线性变换等预处理,以提高图像质量、增强背景与目标的对比度;然后根据谱残差理论对预处理后的图像进行视觉显著性处理,获取包含数字区域的视觉显著图;最后,运用卷积神经网络对显著图进行分类,以区分出数字区域和非数字区域。实验结果表明,该方法可以很好地检测输送带图像上的数字,进而实现对输送带表面故障的定位。A surface fault location of conveyor belt based on saliency and deep convolution neural network was proposed.The method imprints figures on the edge of upper and lower surfaces of conveyor belt,and uses image processing technology to detect the number in belt image,so as to indirectly locate surface fault of the conveyor belt.Firstly,the acquired image of the conveyor belt is preprocessed by Gaussian filtering and gray-scale linear transformation to improve image quality and enhance contrast between the background and the target.Then,visual saliency treatment is conducted to the preprocessed image according to spectral residual theory,and a visual saliency map containing numeric regions is obtained.Finally,saliency map is classified by using the convolution neural network to distinguish digital region from non-digital region.The experimental results show that the method can detect number of conveyor belt image and realize surface fault location of conveyor belt.

关 键 词：输送带 故障定位 神经网络 视觉显著性 图像处理 

分 类 号：TD528.1[矿业工程—矿山机电]