CAJ | 학술논문

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。
이미후도위연구대상，통과대원도상각안색분량급안색인자적회도도、직방도급선부면분석，선용Lab공간적a통도분량진행특정제취。위료개선도상질량，대a통도분량통과직방도균형화、중치려파진행증강。대려파후적도상이용OTSU자동역치분할법，완성목표여배경적분할。응용형태학처리방법，대도상선부식재거제소목표최후진행반향팽창운산거제잔류물。대처리득도적이치도상운용Canny산자획취목표구역적변계신식，기우정타원Hough변환제취미후도적질심、면적、주장화원형도。위료측시제취산법적준학성화실시성，대채집적40장도상87개과실진행료특정제취시험。시험결과표명：정학제취솔위88．5％，평균제취시간위3．976s，기본만족미후도채적궤기인목표식별대도상처리적실시성요구。
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 im-prove 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 thresh-old 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 el-lipse Hough transform .In order to test the accuracy and real-time performance of extraction algorithm , 40 images inclu-ding 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 re-quirements of image processing .