摘要
图像分类技术在农业领域应用广泛,尤其在病害检测和分类方面,相比传统的人工方法更高效和准确。传统的特征融合方法采用固定的加权操作来增强局部特征,并抑制干扰特征的表达,但病害类别图片的差异影响模型泛化能力,导致分类效率和准确率较低。为此,提出一种基于多层自适应特征融合的番茄叶病害识别方法,先通过数据增强算法对数据集进行增强,缓解数据样本量不足、类别不平衡的问题;然后利用特征增强捕捉关键特征,再通过自适应权重的特征融合,以此实现番茄叶病害类别的精准识别。本方法对番茄叶病害图像识别准确率达到99.67%,对比其他InceptionV3、ResNet50等深度网络模型,识别准确率提高2.07%~15.33%。本方法实现对番茄叶病害的图像精准识别,为番茄等农作物病害的识别技术提供思路与方法。
Image classification techniques were widely used in agriculture,especially in disease detection and classification,and were found to be more efficient and accurate than traditional manual methods.Traditional feature fusion methods used fixed weighting operations to enhance local features and suppress the expression of interfering features,but the differences in disease class images affected the generalization ability of the model,resulting in lower classification efficiency and accuracy.In order to address this issue,a tomato leaf disease recognition method based on multi‑layer adaptive feature fusion was proposed in this study.The data set was first enhanced by a data enhancement algorithm to alleviate the problems of insufficient data sample size and category imbalance.Then,feature enhancement was used to capture key features,followed by feature fusion with adaptive weights,resulting in accurate recognition of tomato leaf disease categories.The proposed method achieved a recognition accuracy of 99.67%for tomato leaf disease images,which was an improvement of 2.07%-15.33%compared to other deep network models such as InceptionV3 and ResNet50.Accurate image recognition of tomato leaf diseases was achieved by using this method,which provided ideas and methods for the recognition technology of tomato and other crop diseases.
作者
杨胜英
潘炜垚
雷景生
张淑萍
钱小鸿
Yang Shengying;Pan Weiyao;Lei Jingsheng;Zhang Shuping;Qian Xiaohong(College of Information and Electronic Engineering,Zhejiang University of Science and Technology,Hangzhou,310000,China;Xinjiang Institute of Technology,Akesu,843100,China)
出处
《中国农机化学报》
北大核心
2024年第10期241-246,共6页
Journal of Chinese Agricultural Mechanization
基金
浙江省教育厅科研项目资助(Y202352263)
新疆维吾尔自治区自然科学基金(2022D01C349)。
关键词
数据增强
番茄叶病害
图像分类
特征融合
特征增强
data enhancement
tomato leaf disease
image classification
feature fusion
feature enhancement
