基于BiFormer机制改进的YOLOv5番茄目标检测与应用  

Tomato Object Detection and Application Based on the Improved YOLOv5 Using the BiFormer Mechanism

作　　者：夏少芳 李朝霞 曹博言 隋春荣 XIA Shaofang;LI Zhaoxia;CAO Boyan;SUI Chunrong(School of Mathematics and Information Technology,XingTai University,XingTai,Hebei 054001,China;School of Mathematics and Physics,University of Science and Technology Beijing,Beijing 100083,China)

机构地区：[1]邢台学院数学与信息技术学院,河北邢台054001 [2]北京科技大学数理学院,北京100083

出　　处：《邢台学院学报》2025年第2期106-115,共10页Journal of Xingtai University

基　　金：2023年度河北省科技厅农业科技成果转化资金项目课题:农业生态环境监测与预警系统在果园中的应用与示范(冀科农函[2022]47号);2024年度邢台学院校级科研项目课题:基于视觉检测的草莓采摘机器人相关技术的研究与应用(XTXYYB202414)。

摘　　要：目前果蔬的采摘仍然以人工为主,人工采摘耗时耗力,效率低下。采摘机器人的视觉系统是其获取目标果实信息的关键,对完成采摘动作起着关键性作用。以果蔬园中小番茄为研究对象,研究了自然环境下成熟小番茄的识别与定位,以提高小番茄采摘机器人采摘作业的实时性与鲁棒性。通过不同时段,不同光照,不同遮挡等条件下的小番茄拍摄图片构建数据集,基于YOLOv5对识别模型进行了研究和构建,成功构建了基于BiFormer机制改进的小番茄目标检测模型YOLOv5_tomato,实验结果表明该模型能很好地实现复杂自然环境下小番茄果实目标定位。Presently,the fruit and vegetable picking largely relies on manual labor,characterized by inefficiency due to its time-consuming and labor-intensive nature.The visual perception system of robotic harvesters constitutes a pivotal component,essential for acquiring critical information about targeted fruits, which is fundamental to the execution of harvesting maneuvers. Focusing specifically onsmall tomatoes found within orchard settings, this research delves into the recognition and preciselocalization of ripe small tomatoes under uncontrolled natural conditions. Such efforts aim to significantlybolster both the real-time operational capabilities and robustness of robotic systems dedicated to smalltomato harvesting. To this end, a comprehensive dataset was meticulously assembled, encompassingimages of small tomatoes captured across diverse temporal periods, lighting scenarios, and occlusionconditions. Employing the YOLOv5 framework as the foundational architecture, an exhaustiveinvestigation and subsequent construction of a recognition model were undertaken. This endeavorculminated in the successful development of the YOLOv5_ tomato object detection model, notablyenhanced through the incorporation of the innovative BiFormer mechanism. Rigorous experimentalevaluations have substantiated the model’s efficacy in achieving accurate localization of small tomatofruit targets amidst the complexities inherent to natural environmental settings.

关 键 词：果蔬采摘 YOLOv5 BiFormer机制 小番茄目标检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]