江苏农业学报2025，Vol.41Issue(8)：1538-1552,15.DOI:10.3969/j.issn.1000-4440.2025.08.010
面向水稻杂草识别的高精度图像分类算法
High-precision image classification algorithm for recognition of rice weed
摘要
Abstract
To achieve accurate identification and removal of weeds during the seedling stage in rice fields,this study carried out a systematic study.By collecting and organizing images from real rice field environments,a rice-weed image classi-fication dataset was constructed.Based on this dataset,an innovative and efficient rice-weed image classification algorithm named YOLOv8n-cls-Swift was proposed.During the image feature extraction phase,SwiftFormer was employed to effectively extract discriminative features between rice plants and weeds under complex field conditions.In the classification prediction phase,an efficient weighted classification layer was designed to enable the model to focus more accurately on highly discrimi-native target feature regions,and significantly enhanced its ability to capture distinguishing characteristics.The results demonstrated that,the proposed model achieved a high rec-ognition accuracy.The precision herbicide application sys-tem for rice fields presented in this study can achieve fully automatic identification and removal of weeds,which is ex-pected to play a significant role in reducing pesticide use,improving efficiency,and protecting the environment in rice cultivation.关键词
图像识别/水稻/深度学习/卷积神经网络/图像处理Key words
image recognition/rice/deep learning/convolutional neural network/image processing分类
农业科技引用本文复制引用
杨淑婷,李季,刘正予,马聪,王蓉..面向水稻杂草识别的高精度图像分类算法[J].江苏农业学报,2025,41(8):1538-1552,15.基金项目
宁夏自然科学基金项目(2023AAC03411) （2023AAC03411）
宁夏回族自治区重点研发项目(2023BCF01051、2024BBF01013) （2023BCF01051、2024BBF01013）
