杨长辉,王卓,熊龙烨,刘艳平,康曦龙,赵万华.基于Mask R-CNN的复杂背景下柑橘树枝干识别与重建[J].农业机械学报,2019,50(8):22-30,69.
YANG Changhui,WANG Zhuo,XIONG Longye,LIU Yanping,KANG Xilong,ZHAO Wanhua.Identification and Reconstruction of Citrus Branches under Complex Background Based on Mask R-CNN[J].Transactions of the Chinese Society for Agricultural Machinery,2019,50(8):22-30,69.
摘要点击次数: 1869
全文下载次数: 1085
基于Mask R-CNN的复杂背景下柑橘树枝干识别与重建   [下载全文]
Identification and Reconstruction of Citrus Branches under Complex Background Based on Mask R-CNN   [Download Pdf][in English]
投稿时间:2019-05-29  
DOI:10.6041/j.issn.1000-1298.2019.08.003
中文关键词:  柑橘  采摘机器人  Mask R-CNN  识别  二维重建
基金项目:重庆市重点产业共性关键技术创新专项(cstc2015zdcy-ztzx70003)
作者单位
杨长辉 重庆理工大学
西安交通大学 
王卓 重庆理工大学 
熊龙烨 重庆理工大学 
刘艳平 重庆理工大学 
康曦龙 重庆理工大学 
赵万华 西安交通大学 
中文摘要:为了获取自然环境下完整柑橘果树枝干信息,以引导采摘机器人进行避障采摘作业,提出了一种基于Mask R-CNN模型与多参数变量约束的柑橘果树枝干识别与重建方法。该方法采用网格化的标记方式对果树枝干进行标记,完成了对柑橘果树枝干的小区域识别;然后对该模型获得的离散mask进行最小外接矩处理,以获得更精确的目标区域;接着利用多参数变量约束完成同一枝干mask(掩码)的划分;最后为了使重建枝干更符合实际枝干的生长姿态,以及完善未识别区域的检测,对同一枝干mask中心点进行了四次多项式拟合。实验结果表明,模型在测试集下的平均识别精确率为98.15%,平均召回率为81.09%,果树单条枝干平均拟合误差为11.47%,果树枝干整体平均重建准确率为88.64%。
YANG Changhui  WANG Zhuo  XIONG Longye  LIU Yanping  KANG Xilong  ZHAO Wanhua
Chongqing University of Technology;Xi’an Jiao Tong University,Chongqing University of Technology,Chongqing University of Technology,Chongqing University of Technology,Chongqing University of Technology and Xi’an Jiao Tong University
Key Words:citrus  picking robot  Mask R-CNN  recognition  2D reconstruction
Abstract:In the process of citrus harvesting, it is necessary to obtain information about branches and trunks of fruit trees for obstacle avoidance. In natural environment, problems such as random growth posture, different shapes and blocked branches and trunks often arise. In order to complete the acquisition of complete information of branches, the small area recognition of citrus fruit branches was completed by grid marking. The precision rate of the training model under the test set was 98.15% and the average recall rate was 81.09%, and the marker formula could still achieve better recognition results. Because the identified small areas were discrete and discontinuous, it was necessary to divide and sort the discrete areas in order to reconstruct the branches and trunks of citrus trees. At the same time, in order to solve the problems of too many background areas in Mask R-CNN model recognition frame and the recognition frame can not rotate with the growth of branches, the discrete mask obtained from Mask R-CNN model was processed with minimum external moment, and the rectangular border with minimum external moment was used to replace the recognition frame of the original model. Secondly, through the statistical analysis of the position information such as angle and distance between the centerlines of adjacent recognition frames after processing, it was found that there were constraints on the parameters such as angle and distance between centerlines. Therefore, it was proposed to use multi-parameter variable constraints to complete the division of identical recognition frames, in order to reconstruct the branches more in line with the actual growth posture of the branches and improve the ignorance. In the detection of different regions, the center point of the identical trunk recognition frame was fitted by quadratic polynomial, and the fitting error was 11.47%. Finally, the experimental results showed that the citrus tree branch reconstruction accuracy rate was 8864%. This method can provide a basis for the robot to avoid obstacles safely.

Transactions of the Chinese Society for Agriculture Machinery (CSAM), in charged of China Association for Science and Technology (CAST), sponsored by CSAM and Chinese Academy of Agricultural Mechanization Science(CAAMS), started publication in 1957. It is the earliest interdisciplinary journal in Chinese which combines agricultural and engineering. It always closely grasps the development direction of agriculture engineering disciplines and the published papers represent the highest academic level of agriculture engineering in China. Currently, nearly 8,000 papers have been already published. There are around 3,000 papers contributed to the journal each year, but only around 600 of them will be accepted. Transactions of CSAM focuses on a wide range of agricultural machinery, irrigation, electronics, robotics, agro-products engineering, biological energy, agricultural structures and environment and more. Subjects in Transactions of the CSAM have been embodied by many internationally well-known index systems, such as: EI Compendex, CA, CSA, etc.

   下载PDF阅读器