基于数据增强的语义一致番茄叶病识别方法
Semantic Consistent Tomato Leaf Disease Recognition Method Based on Data Augmentation
丁雪莲1
作者信息
- 1. 内蒙古财经大学计算机信息管理学院,呼和浩特 010070
- 折叠
摘要
针对番茄叶片病害识别中缺乏训练图像的问题,提出一种基于数据增强的语义一致番茄叶病识别方法.首先,设计数据增强模块,对数据集进行有效扩充.然后,定义深度特征提取模块,捕捉图像中丰富的非线性语义信息.同时,为了避免数据增强过程中的语义漂移,设计语义相关最大化模块,增强原始数据和增强数据的语义相关性.最后,定义番茄叶病识别模块,实现番茄叶片病害识别.实验结果表明,相比于其他8个基线方法,本文方法在识别番茄叶片病害的准确率上达到了更优.
Abstract
In response to the challenge of insufficient training images for tomato leaf disease rec-ognition,an innovative semantic-consistent tomato leaf disease recognition method is proposed based on data augmentation.Firstly,a data augmentation module is designed to effectively expand the dataset.Next,a deep feature extraction module is defined to capture rich nonlinear semantic infor-mation from the images.To mitigate semantic drift during the data augmentation process,a seman-tic relevance maximization module is introduced to enhance the semantic consistency between origi-nal and augmented data.Finally,a leaf disease recognition module is defined to achieve tomato leaf disease identification.Experimental results indicate that our approach outperforms eight baseline methods in terms of accuracy.
关键词
番茄叶病识别/数据增强/深度特征/语义相关性Key words
tomato leaf disease recognition/data augmentation/deep feature/semantic correlation引用本文复制引用
基金项目
内蒙古自治区科技计划(2022YFSJ0018)
出版年
2024
NETL
NSTL
万方数据