基于YOLOv7和YCrCb的火龙果识别与分割方法  

Identification and Segmentation Method of Dragon FruitBased on YOLOv7 and YCrCb

作　　者：罗陈迪 李文涛 商枫楠 肖明玮 陈桥 欧阳春凡 周学成 Luo Chendi;Li Wentao;Shang Fengnan;Xiao Mingwei;Chen Qiao;Ouyang Chunfan;Zhou Xuecheng(Guangdong Provincial Key Laboratory of Agricultural Artificial Intelligence(GDKL-AAI),Guangzhou 510642,China;Key Laboratory of Key Technology on Agricultural Machine and Equipment,Ministry of Education,South China Agricultural University,Guangzhou 510642,China;College of Engineering/Guangdong Engineering Research Center for Agricultural Aviation Application(ERCAAA),South China Agricultural University,Guangzhou 510642,China)

机构地区：[1]广东省农业人工智能重点实验室,广州510642 [2]华南农业大学南方农业机械与装备关键技术教育部重点实验室,广州510642 [3]华南农业大学工程学院/广东省农业航空应用工程技术研究中心,广州510642

出　　处：《农机化研究》2025年第7期59-64,共6页Journal of Agricultural Mechanization Research

基　　金：广东省科技计划项目(2021B1212040009)。

摘　　要：现阶段火龙果主要通过人工采摘,采摘不及时可能会导致果实衰老,造成果实品质损失,且不利于贮藏。研究火龙果采摘机器人,实现自动化采摘是解决上述问题的重要举措,而火龙果图像识别与分割是自动采摘的关键技术环节。为此,提出了一种基于YOLOv7和YCrCb色彩空间相结合的算法,对复杂背景下的火龙果进行识别与分割,YOLOv7网络的检测框内通过引入YCrCb色彩空间,联合OTSU阈值分割算法和形态学操作等方法实现果实与背景的分割。为了验证YOLOv7网络的性能,与Faster R-CNN网络进行比较,结果表明:在相同试验条件下YOLOv7平均检测精度为98.82%,提高了6.81%;F1值为0.95,提高了0.22;此外,通过YCrCb可以较好地分割火龙果,平均用时约108 ms。At the current stage,dragon fruits are primarily harvested manually.Delayed harvesting can result in less freshness of fruits,leading to a loss in fruit quality and compromising their storage potential.To address these issues,the research on developing dragon fruit harvesting robots for automation has become an important initiative.Image recognition and segmentation of dragon fruits play a crucial role in achieving automated harvesting.For this,proposed an algorithm that combines YOLOv7 and the YCrCb color space for the recognition and segmentation of dragon fruits in the complex background.The algorithm utilized the YCrCb color space in conjunction with methods such as the OTSU threshold segmentation algorithm and morphological operations to segment the fruit from the background within the detection boxes of the YOLOv7 network.In order to evaluate the performance of the YOLOv7 network,a comparison was made with the Faster R-CNN network under the same experimental conditions.The results showed that the average detection accuracy of YOLOv7 improved by 6.81%to 98.82%,and the F 1 score increased by 0.22 to 0.95.Additionally,the YCrCb color space allowed for effective segmentation of dragon fruits with an average processing time of approximately 108 ms.

关 键 词：火龙果 YOLOv7 图像分割 YCrCb OTSU算法 

分 类 号：S225[农业科学—农业机械化工程]