基于计算机视觉的香菇缺陷检测

传统人工检测鲜香菇表面缺陷方法通常效率较低,且易造成视觉疲劳,不能满足现代化工业需求。本研究尝试采用计算机视觉技术检测鲜香菇表面缺陷。首先,香菇样本RGB图像被获取,抽取B分量图像构建掩模用于G分量图像去背景。然后,去背景后的G分量图像进行边缘亮度补偿及缺陷提取。随后,对缺陷标记、提取特征参数及参数选择。为了避开成像系统及环境光的干扰,试验选用缺陷区域总面积与香菇图像总面积的比值作为识别正常香菇与缺陷菇的衡量指标。最后,一个全局阈值0.0035被用于所有被研究样本。结果表明,该识别算法识别正常菇和缺陷菇准确率分别为94%和97.3%,所有样本分类精度达到96.5%。

Defects Detection of Lentinus Edodes Surface Based on Computer Vision Technology

The traditional artificial method of detecting defects on entinus edodes surface is normal less efficiency and easily vision tiredness.Therefore,it can not satisfy the demand of the modern industry production.This research attempted to detect entinus edodes surface defects based on computer vision technology.Firstly,RGB images of samples were acquired and B component images were extracted to build mask for removing G component images background.Then,edge light compensation and defects extraction method were...