基于改进YOLOv8模型的葡萄花穗和幼果分类检测方法及试验

打开文本图片集

中图分类号：TP391.4；S24 文献标识码：A 文章编号：2095-5553（2026）03-0081-09

Abstract：Inorder toimprove theeficiencyandqualityof grape plant protectionrobots andachieve precise positioning of therobot’s front-end spraying mechanism，this study proposesa grape inflorescenceandfruitletrecognitionmethod YOLOv8—FCD based onan improved YOLOv8 model，and verifies itthrough field experiments.TheC2f_Faster module based on PConv was introduced to reduce the model's parameter volume and computational load.And then the CARAFE was used toreplaceasa fusion-network up-sampling module to enhance feature extraction capabilities.Finally，SEAM Was introducedtoredesignthe detection headandobtaintheDetect_SAMdetectionmodule，achieving thegoal of further improving detection accuracy. The experimental results indicate that the precision rate，recall rate，and mAP value of the improved YOLOv8—FCD model are 93.7% ， 87.3% ，and 94.6% ，respectively.Compared with the original model， the precision rate，recall rate，and mAP value of the model have increased by 8.2% ， 2.2% ，and 2.6% ，respectively，and the model weight has decreased by 14.29% . The model was deployed to the grape plant protection spray test prototype for verification tests. The results showed that the inference time of the upper computer accounted for about 24.95% of the complete operation time，and the spray success rate was 85% .It was verified that the model could provide technical support for grape intelligent spray plant protection operations. Keywords：grape；image processing；flower clusters；deep learning；object detection； plant protection robot

0 引言

我国葡萄产业已经形成集约化和规模化，是全球第二大葡萄生产国和全球第一大鲜食葡萄生产国[1]。（剩余14545字）