CAJ | 학술논문

为实现香蕉形状的识别，文章提出了一种基于计算机视觉和椭圆傅里叶描述子的形状识别方法。先由专家挑选果指果形为弯、微弯、末端直和直的香蕉共107枚，再获取预处理后图像的边界链码并提取其椭圆傅里叶系数，通过比较原形状和重建形状来选取合适的系数个数，然后用主成分分析对非负系数进行降维，最后进行模糊C均值聚类。研究结果表明，可以将该机器识别方法应用于香蕉形状的识别，为实现香蕉及其深加工产品的增值提供了一种技术手段，并为进一步实现香蕉果把和果穗的形状识别奠定了基础。
위실현향초형상적식별，문장제출료일충기우계산궤시각화타원부리협묘술자적형상식별방법。선유전가도선과지과형위만、미만、말단직화직적향초공107매，재획취예처리후도상적변계련마병제취기타원부리협계수，통과비교원형상화중건형상래선취합괄적계수개수，연후용주성분분석대비부계수진행강유，최후진행모호C균치취류。연구결과표명，가이장해궤기식별방법응용우향초형상적식별，위실현향초급기심가공산품적증치제공료일충기술수단，병위진일보실현향초과파화과수적형상식별전정료기출。
In this paper , a method based on computer vision and elliptic Fourier descriptor ( EFD) was developed to recognize the banana shape .The number of banana with shapes of the curved , slightly curved , end-straight and straight was 107 , and the bananas were previously indentified by the experts .Subsequently , the boundary chain codes of the pre-processed images of these bananas were obtained for extracting the elliptic Fourier coefficients .The suitable number of coefficients was determined by comparing the original and reconstructed banana shape , and then the principal component analysis was used to reduce the number of non-negative coefficients .Finally, the classifica-tion was achieved by the fuzzy C-means clustering .The results demonstrated that this machine method could be ap-plied for banana shape identification in practice .This study provided a technological means for the value increase of banana and its secondary products , and also might be the basis of realizing the shape recognition of banana hand and bunch in the future .