CAJ | 학술논문

研究提出了一种基于机器视觉的病害和裂纹的识别方法. 在H 分量图中,依据半干枣在病害和非病害区域色调值差异提取病害区域,以提取的病害区域与枣表面积的比作为阈值确定较高的病害面积识别精度,可正确识别的感兴趣病害面积为16.87mm2 ,占枣投影面积的3.3%. 为进一步提高在该病害面积识别精度的正确率,依据已确定的病害面积比阈值,将病害面积比值二值化,结合红枣区域颜色特征值 H 的均值和均方差,用SVM方法建立枣病害的识别模型,训练集和测试集的识别正确率分别为95.77%和95.79%. 在 I 分量图中,对红枣区域进行Otsu ' s 阈值分割、图像局部属性统计和形态学处理,提取裂纹二值图像,依据裂纹图像不变距方法建立裂纹识别模型,训练集和测试集的识别正确率分别为94.90%和94.55%.
연구제출료일충기우궤기시각적병해화렬문적식별방법. 재H 분량도중,의거반간조재병해화비병해구역색조치차이제취병해구역,이제취적병해구역여조표면적적비작위역치학정교고적병해면적식별정도,가정학식별적감흥취병해면적위16.87mm2 ,점조투영면적적3.3%. 위진일보제고재해병해면적식별정도적정학솔,의거이학정적병해면적비역치,장병해면적비치이치화,결합홍조구역안색특정치 H 적균치화균방차,용SVM방법건립조병해적식별모형,훈련집화측시집적식별정학솔분별위95.77%화95.79%. 재 I 분량도중,대홍조구역진행Otsu ' s 역치분할、도상국부속성통계화형태학처리,제취렬문이치도상,의거렬문도상불변거방법건립렬문식별모형,훈련집화측시집적식별정학솔분별위94.90%화94.55%.
Diseases and cracks are the common defects of red dates and they severely reduce the quality of red dates . The objective of this study was to determine the effectiveness of a computer vision system with RGB color camera in detec -ting the diseases and surface cracks in red dates .Firstly , on the basis of the difference in the tone value between the dis-eased and non-diseased areas in the H diagram , diseased area was extracted , and the extracted disease area to total sur-face area ratio was used as the threshold to achieve a high precision in identifying the diseased area .The test results of 163 diseased red dates and 500 non-diseased dates showed that more than 16 .87 mm2 diseased area could be correctly identified , accounting for 3 .3%of the projected area of a red date .The rates of correct recognition for the training set and the test set were 92 .60% and 91 .58%, respectively .To further improve the accuracy , the extracted diseased area to the surface area ratio was converted to the binary format .Combining with the mean and variance of color features of the red dates, an SVM ( support vector machine ) model was developed to detect red date diseases .The correct detection rate was 95.77 %for the training data set and 95.79%for the test data set.In the I diagram, Otsu's threshold method was firstused to segment the regions on date surface , and then statistical and morphological methods were used to segment the crack regions and generate binary images .Using the invariant of cracks in the crack binary images , a crack recognition model was established .The adequacy of the model was tested on a data set of 500 samples , including 148 cracked dates and 352 non-cracked dates.For training data set, the detection rate was 94.9%.For the test data set, the detection rate was 94 .55%.The results showed that it was feasible to use the machine vision for disease and crack identification of semi-dried dates .The method could potentially be used for on-line detection of external quality of semi-dried dates .