摘要
在传统的作物病害识别的深度学习模型中,存在检测精度与效率不高的问题。针对上述问题提出一种轻量化的改进型MobileNet V2模型CA-MobileNet V2(coordinate attention),在提升检测精度的同时,部署在移动端便于种植者使用。在MobileNet V2中嵌入坐标注意力模块,提升模型的精度;加入TanhExp激活函数,加速模型收敛,增强模型的鲁棒性和泛化性;将模型部署到移动端APP中,使模型具有良好的可视化应用效果。在PantifyDr和Turkey-PlantDataset数据集上的对比实验结果表明,CA-MobileNet V2具有精度高和轻量化的优势。
In the traditional deep learning models for crop disease identification,there are problem of low detection accuracy and efficiency.A lightweight and improved MobileNet V2 model,namely CA-MobileNet V2(coordinate attention),was proposed for the above problem,which was easy to use by growers while improving the detection accuracy and deploying on mobile terminal.The lightweight coordination attention module was embedded in MobileNet V2 to improve accuracy with almost no computational overhead.TanhExp activation function was added for the lightweight network to accelerate model convergence and enhance model robustness and generalization.The model was deployed to the mobile APP,so that the model had better visual application effects.The results of comparison experiments on PantifyDr and Turkey-PlantDataset datasets show that CA-MobileNet V2 has the advantages of high accuracy and light weight.
作者
陈洋
张欣
陈孝玉龙
林建吾
蔡季桐
CHEN Yang;ZHANG Xin;CHEN Xiao-yu-long;LIN Jian-wu;CAI Ji-tong(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China;College of Agriculture,Guizhou University,Guiyang 550025,China)
出处
《计算机工程与设计》
北大核心
2024年第2期484-490,共7页
Computer Engineering and Design
基金
国家自然科学基金项目(61865002)
国家重点研发计划重点专项基金项目(2021YFE0107700)。
关键词
农作物病害
深度学习
卷积神经网络
轻量化
坐标注意力
激活函数
移动端部署
crop diseases
deep learning
convolutional neural networks
lightweight
coordinate attention
activation function
mobile deployment