基于机器视觉的猕猴桃特征参数提取

Characteristic Parameters Extraction of Kiwifruit Based on Machine Vision

以猕猴桃为研究对象,通过对源图像各颜色分量及颜色因子的灰度图、直方图及线剖面分析,选用Lab空间的a通道分量进行特征提取。为了改善图像质量,对a通道分量通过直方图均衡化、中值滤波进行增强。对滤波后的图像利用OTSU自动阈值分割法,完成目标与背景的分割。应用形态学处理方法,对图像先腐蚀再去除小目标最后进行反向膨胀运算去除残留物。对处理得到的二值图像运用Canny算子获取目标区域的边界信息,基于正椭圆Hough变换提取猕猴桃的质心、面积、周长和圆形度。为了测试提取算法的准确性和实时性,对采集的40张图像87个果实进行了特征提取试验。试验结果表明:正确提取率为88.5%,平均提取时间为3.976s,基本满足猕猴桃采摘机器人目标识别对图像处理的实时性要求。

Kiwifruit was used as the research object,a component of Lab space was choosed for feature extraction after analysing grayscale,histogram and line profile of each color component and color factor of source image.In order to improve visual effect of the image,a channel component was enhanced through histogram equalization and median filter.In orde to partition the target and the background,the filtered image was managed by the method of Otsu automatic threshold segmentation,the target and the background segmentation.In order to remove residues,image was processed through morphological principles,first to remove the image in the small target,corrosion,corrosion of the image again to remove the small target,and finally reverse-expansion methods.the processed binary image was first used Canny operator for edge information of the target area,then kiwi＇s center of mass,area,circle and circularity were extracted based on ellipse Hough transform.In order to test the accuracy and real-time performance of extraction algorithm,40 images including 87 fruits was used for feature extraction experiment,experimental results show that the correct extraction yield was 88.5%,the average time of 3.976 seconds,basically meet the kiwifruit picking robot target recognition real-time requirements of image processing.