摘要
针对自然场景下生长期内树上未成熟果实的自动探测与大小计算问题,提出了一种基于改进分水岭和凸包理论的自然场景下未成熟苹果识别与直径计算方法。该方法首先对灰度图像进行形态学重构后进行边缘检测,再利用合并局部极小值点分水岭分割方法从粘连区域中提取目标果实,并结合基于凸包理论的真轮廓提取和圆拟合方法,实现目标果实圆拟合直径的自动测量。计算结果与人工测量结果进行对比试验,结果表明:在不考虑扁平型目标果的情况下,该方法的直径计算均方根误差最小值为1.91 mm,均值为2.27 mm,误差范围在品质评定等级差(5 mm)以内,具有较好的推广应用价值。研究结果为生长期内果实的大小监测提供参考。
In agriculture field, the knowledge about the size of immature fruit in its growing period could be helpful to the aspects such as precision fertilization and irrigation, training and pruning, yield estimation and harvest stage determination. Thus, it can improve the fruit yield and quality. However, the most important premise is the fruit detection and the size calculation in natural condition. The color of fruits on tree is very similar with their surrounding objects, so that the recognition and size calculation is very difficult in natural condition. This paper proposed an algorithm of diameter measurement for immature apple based on the morphological reconstruction, the watershed and convex hull theory. Firstly, the images were acquired using the JV205 cameras which were installed in front of the apple trees with the distance of about 1-2 m, whose resolution was 4 608 pixels × 3456 pixels; and acquisition time was from June 1th, 2014 to September 30 th, 2014. Before preprocessing, the images were cropped into 1400 pixels × 1100 pixels to get the scope of the target apples. In order to enhance the visibility of the target area, the morphological reconstruction operation was used before the edge detection. Secondly, the rough contour of target fruit was extracted by the dilation and erosion operation using micro-disk structure element with a radius of two pixels. After the above image preprocessing, excessive local minimum points, which were caused by occluded leaves and uneven illumination, were merged using the bresenham algorithm. And then the overlapping target fruits were segmented by distance transform and improved watershed algorithm. Later, the contours of segmented target region were tracked and the continuous smooth contour curve was extracted using convex hull on the edge of the segmentation images. Finally, the entire circle was reconstructed using the three different points on the reserved true target contour based on the circle feature principle. Thus, the center and radius parameters were estim
出处
《农业工程学报》
EI
CAS
CSCD
北大核心
2014年第23期207-214,共8页
Transactions of the Chinese Society of Agricultural Engineering
基金
国家自然科学基金青年基金项目(31301238)
北京市自然科学基金资助项目(4132027)
北京市农林科学院国际合作基金项目(GJHZ2013-4)
欧盟FP7项目(PIRSES-GA-2013-612659)
关键词
果实
图像处理
形态学
分水岭
凸包理论
直径测量
fruits
image processing
morphology
watershed
convex hull
diameter measurement
