双通道YOLOv8s棉花地膜识别与定位系统  

Dual-channel YOLOv8s Cotton Mulch Film Identification and Localization System

作　　者：董芙楠 陈东胜 潘江如 董恒祥 DONG Funan;CHEN Dongsheng;PAN Jiangru;DONG Hengxiang(School of Control Engineering,Xinjiang College of Engineering,Urumqi 830023,China;Kuitun Yinli Cotton Oil Machinery Co.,Ltd.,Kuitun 833200,China;Party and Government Office,Xinjiang Hotan College,Hotan 848000,China;School of Automotive Engineering,Xinjiang Transportation Vocational Technical College,Urumqi 831401,China)

机构地区：[1]新疆工程学院控制工程学院,新疆乌鲁木齐830023 [2]奎屯银力棉油机械有限公司,新疆奎屯833200 [3]新疆和田学院党政办,新疆和田848000 [4]新疆交通职业技术学院汽车工程学院,新疆乌鲁木齐831401

出　　处：《北华大学学报(自然科学版)》2025年第3期402-412,共11页Journal of Beihua University(Natural Science)

基　　金：中央引导地方科技发展专项资金项目(ZYYD2023B19)。

摘　　要：在机械采摘棉花过程中,采棉机常将残留地膜混入皮棉,影响棉花品质。提出一种基于双通道YOLOv8s的棉花地膜识别与定位方法,旨在提高地膜清除效率。改进原有工艺流程:在锯齿皮清机和集棉机间集成了ZED双目相机以实时采集图像信息;将采集的图像输入改进YOLOv8s模型中识别,并利用双目相机获取的位置信息进行空间定位。YOLOv8s模型改进:在原有模型的13层引入ECA注意力模块;引入小目标检测层,提高对小目标标定的能力;设计一种新的双通道注意力结构,提高模型的泛化能力。通过消融试验验证本文改进方法的有效性,试验结果表明:双通道YOLOv8s模型在自制棉花地膜数据集上达到了90.92%的mAP@0.5,比改进前的YOLOv8s提高了10.93%。利用ZED双目相机测距功能,获取空间三维坐标,能够快速、准确地识别与定位棉花地膜,为下位机精准去除棉花中的地膜提供了依据。In the process of mechanical picking cotton,cotton picking machine often mix residual mulch film into the lint,affecting the quality of cotton.A dual-channel YOLOv8s-based method for cotton mulch film identification and localization was proposed,aiming to improve the efficiency of mulch removal.The original process was improved by integrating a ZED stereo camera system between the serrated cotton cleaner and the cotton collector to collect real-time image information.The collected images were input into the improved YOLOv8s model for recognition,using position information acquired by stereo cameras for spatial localization.Improvement of the YOLOv8s model:an ECA attention module was introduced into the original model’s 13 layers;a small target detection layer was introduced to enhance the ability to identify small targets;a new dual-channel attention structure was designed to improve the model’s generalization ability.The effectiveness of the improved method was verified by ablation tests.The experimental results show that the dual-channel YOLOv8s model achieved an mAP@0.5 of 90.92%on the self-built cotton mulch film dataset,an improvement of 10.93%over the previous YOLOv8s model.The spatial three dimensional coordinates were obtained by using the ZED stereo camera’s distance measurement function.This method can quickly and accurately recognize and locate cotton mulch film,providing a basis for the lower-level machine to precisely remove the mulch film from the cotton.

关 键 词：双通道 YOLOv8s 棉花地膜 识别与定位 ZED双目相机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]