| 摘要: |
| 为提高运动苹果的采后分选精度，设计一种基于苹果朝向的最大横切面直径检测方法。采集苹果在传送杆上多位置、多姿态RGB图像，利用张氏标定法进行图像畸变校正，通过颜色分割和形态学处理方法提取苹果区域；基于R通道分量的梯度变化获取苹果果梗/花萼区中心，通过对称性判断苹果朝向。计算垂直于朝向的最大直径，将其作为最大横切面直径；针对相机投影角增大产生的投影失真问题，构建投影校正方程以校正检测结果。结果表明：1）对60个苹果（富士和黄元帅各30个）的720幅图像进行直径检测，以单张图像检测结果在±5 mm内作为检测准确的判断标准，所提出的投影校正算法将单张图像的直径检测准确率从65.3%提升至95.3%。2）对每个苹果的12张图像的直径检测结果求平均值作为最终果径，准确率达100%，其中，86.7%的富士和93.3%的黄元帅检测直径在±2 mm以内，均方根误差分别为1.5 mm和1.1 mm。总体结果表明，本研究提出的最大横切面直径检测方法能够精确检测苹果果径，可为苹果大小分级提供技术支持，也为近球形的果蔬分选提供了有效参考。 |
| 关键词: 苹果采后分选 苹果朝向 最大横切面直径 投影校正 |
| DOI：10.11841/j.issn.1007-4333.2024.05.12 |
| 投稿时间：2023-10-09 |
| 基金项目:国家重点研发计划（2023YFE0122600）；世界顶尖涉农大学国际合作交流种子基金（15052001） |
|
| Maximum cross‐sectional diameter detection method based on the orientation of apples |
|
YU Jiangfan1, ZHOU Long2, ZHANG Zhao1, JIANG Zuochen3, LI Yunxia1, RUI Zhaoyu1, ZHANG Man1, LI Xu4, YUAN Dongdong5
|
| (1.Key Laboratory of Smart Agriculture System Integration of Ministry of Education， China Agricultural University， Beijing 100083， China;2.College of Computer and Artificial Intelligence， Changzhou University， Changzhou 213164， China;3.College of Mechanical and Electrical Engineering， China Jiliang University， Hangzhou 310018， China;4.College of Information Engineering， Tarim University， Alar 843300， China;5.Sweet Fruit Company Limited， Suqian 223725， China) |
| Abstract: |
| To enhance the postharvest sorting precision of moving apples， a method for detecting the maximum cross-section diameter was devised based on the orientation of the apples.The RGB images of apples in various positions were captured and their orientations on the conveyor were acquired. Image distortion correction was performed using Zhang’s calibration method， and apple areas were isolated through color segmentation and morphological processing. The center of the apple peduncle/calyx region was determined using the gradient change of the R channel component， and apple orientation was assessed based on symmetry. The maximum diameter perpendicular to the orientation was then calculated as the maximum cross-sectional diameter. To address the projection distortion resulting from the increased camera projection angle， a projection correction equation was formulated to rectify the detection outcomes. The results reveal that： 1） Diameter detection was conducted on 720 images of 60 apples （Fuji and Yellow Marshal each have 30）. Single image detection accuracy， defined within ±5 mm， was employed as the criterion. The proposed projection correction algorithm significantly increased the accuracy from 65.3% to 95.3%. 2） The diameter detection results of 12 images per apple were averaged to determine the final fruit diameter， which achieved a 100% accuracy rate. Notably， the 86.7% of Fuji and 93.3% of Yellow Marshal exhibited a detection diameter within ±2 mm with root mean square errors of 1.5 mm and 1.1 mm， respectively. Overall， the method proposed in this study accurately detects the diameter of apple fruits. It can offer valuable technical support for apple size grading and be used as a pertinent reference for sorting near-spherical fruits and vegetables. |
| Key words: apple postharvest sorting apple orientation maximum cross-sectional diameter projection correction |
