基于TransUnet的田间杂草分割研究  

Research on field weed segmentation based on TransUnet

作　　者：高皓章 唐友 辛鹏[1] 朱国东 Gao Haozhang;Tang You;Xin Peng;Zhu Guodong(Jilin Institute of Chemical Technology,Jilin 132022,China;Jilin University of Agricultural Science and Technology,Jilin 132101,China;Faw East Mechanical Shock Absorber Co.,Ltd.,Changchun 130001,China)

机构地区：[1]吉林化工学院,吉林吉林132022 [2]吉林农业科技学院,吉林吉林132101 [3]一汽东机工减振器有限公司,吉林长春130001

出　　处：《无线互联科技》2023年第15期100-103,共4页Wireless Internet Technology

基　　金：吉林省科技发展计划项目,项目名称:基于数据挖掘技术的全基因组选择方法研发及云计算平台体系构建,项目编号:YDZJ202201ZYTS692。

摘　　要：农业生产是人类生存的基础,而农作物的生长过程中总是受到杂草的影响,造成农作物的减产,因此实现田间杂草的智能分割具有重要意义。文章采用TransUnet算法模型进行田间杂草分割,旨在提高分割的准确性和效率。TransUnet是一种新型的深度学习网络。该模型的训练使用了公开数据集,并结合迁移学习的思想对模型进行了优化。本实验通过测试和评估,发现经过迁移学习的TransUnet模型在田间杂草的分割任务中表现突出,其像素准确率可以达到97%以上。此外,文章还对模型进行了可视化分析,证实了模型的实用价值和应用前景。综上所述,TransUnet模型对于杂草的分割有积极效果。希望本研究成果可以为计算机视觉技术在农业行业的应用提供更多的思路和借鉴。Agricultural production is the basis of human survival,and the growth of crops are always affected by weeds,resulting in crop production reduction.Therefore,it is of great significance to realize the intelligent segmentation of weeds in the field.In this study,the TransUnet algorithm model was used for weed segmentation in the field,aiming to improve the accuracy and efficiency of segmentation.TransUnet is a new type of deep learning network.The training of the model uses the public data set,and the model is optimized by combining the idea of transfer learning.In this experiment,through testing and evaluation,it is found that the TransUnet model after transfer learning performs outstanding in the field weed segmentation task,and its pixel accuracy can reach more than 97%.In addition,the visual analysis of the model is carried out,which confirms the practical value and application prospect of the model.In summary,the TransUnet model has a positive effect on weed segmentation.It is hoped that the results of this research can provide more ideas and references for the application of computer vision technology in the agricultural industry.

关 键 词：杂草分割 语义分割 机器视觉 深度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]