摘要
在设施农业种植环境下,番茄花朵存在叶片遮挡、花朵堆叠、背景环境复杂等现象,导致番茄花朵检测模型的准确率较低,限制了授粉机器人的智能化发展。针对该问题,本文提出了一种基于改进YOLOv5s模型的番茄花朵目标检测方法,通过将原有模型中的C3模块替换为C2f_ScConv模块、引入LSKA注意力机制、添加ADown下采样结构模块等对原始YOLOv5s模型结构进行改进。最后,将改进前后的模型进行训练和检测。改进后的YOLOv5s番茄花朵识别模型能够有效的识别出重叠遮挡、小目标的番茄花朵,其mAP值相比于初始模型提升了6.3个百分点,召回率提升了5.6个百分点,准确度提升了6.7个百分点,拥有更强的特征提取能力和鲁棒性。该模型可以满足设施环境中授粉机器人对番茄花朵的识别精度需求。
In the facility agriculture planting environment,tomato flowers have phenomenons such as leaf obstruction,flower stacking,and complex background environment,which lead to low accuracy of tomato flower detection models and limit the intelligent development of pollination robots.To address this issue,this paper proposed a tomato flower object detection method based on an improved YOLOv5s model.The original YOLOv5s model structure was improved by replacing the C3 module with the C2f_ScConv module,introducing the LSKA attention mechanism,and adding the ADown downsampling structure module.Finally,the models were trained and tested before and after improvement.The improved YOLOv5s tomato flower recognition model could effectively recognize tomato flowers with overlapping occlusion and small targets.Compared with the initial model,its mAP value had increased by 6.3 percentage points,recall rate had increased by 5.6 percentage points,accuracy had increased by 6.7 percentage points,and it has stronger feature extraction ability and robustness.This model could meet the recognition accuracy requirements of pollination robots for tomato flowers in facility environments.
作者
朱婷倩
张华
陈丰
张运来
李娜
吴镛
苏祥祥
ZHU Tingqian;ZHANG Hua;CHEN Feng;ZHANG Yunlai;LI Na;WU Yong;SU Xiangxiang(College of Mechanical Engineering,Anhui Science and Technology University,Fengyang 233100,China)
出处
《安徽科技学院学报》
2025年第1期60-69,共10页
Journal of Anhui Science and Technology University
基金
安徽省科技特派员农业物质技术装备领域揭榜挂帅项目(2022296906020001)
安徽省高校自然科学研究重大项目(KJ2021ZD0110)。
