基于无人机遥感和深度学习的葡萄卷叶病感染程度诊断方法  

Diagnosis of Grapevine Leafroll Disease Severity Infection via UAV Remote Sensing and Deep Learning

作　　者：刘易雪 宋育阳[4] 崔萍 房玉林[4] 苏宝峰[1,2,3] LIU Yixue;SONG Yuyang;CUI Ping;FANG Yulin;SU Baofeng(College of Mechanical and Electronic Engineering,Northwest A&F University,Yangling 712100,China;Key Labo‐ratory of Agricultural Internet of Things,Ministry of Agriculture and Rural Affairs,Yangling 712100,China;Shaanxi Key Laboratory of Agriculture Information Perception and Intelligent Service,Yangling 712100,China;College of Enol‐ogy,Northwest A&F University,Yangling 712100,China;Ningxia Helan Mountain East Foothill Wine Industry Park Management Committee,Yinchuan 750002,China)

机构地区：[1]西北农林科技大学机械与电子工程学院,陕西杨凌712100 [2]农业农村部农业物联网重点实验室,陕西杨凌712100 [3]陕西省农业信息感知与智能服务重点实验室,陕西杨凌712100 [4]西北农林科技大学葡萄酒学院,陕西杨凌712100 [5]宁夏贺兰山东麓葡萄产业园区管理委员会,宁夏银川750002

出　　处：《智慧农业（中英文）》2023年第3期49-61,共13页Smart Agriculture

基　　金：宁夏回族自治区重点研发计划项目(2021BEF02017)。

摘　　要：[目的/意义]葡萄卷叶病是一种严重影响葡萄产量和品质的病害。然而,葡萄卷叶病感染程度类别之间存在严重的数据不平衡,导致无人机遥感技术难以进行精确的诊断。针对此问题,本研究提出一种结合细粒度分类和生成对抗网络(Generative Adversarial Network,GAN)的方法,用户提高无人机遥感图像中葡萄卷叶病感染程度分类的性能。[方法]以蛇龙珠品种卷叶病识别诊断为例,使用GANformer分别对每一类的葡萄园正射影像的分块图像进行学习,生成多样化和逼真的图像以增强数据,并以Swin Transformer tiny作为基础模型,提出改进模型CA-Swin Transformer,引入通道注意力机制(Channel Attention,CA)来增强特征表达能力,并使用ArcFace损失函数和实例归一化(Instance Normalization,IN)来改进模型的性能。[结果和讨论]GANformer可以生成FID score为93.20的蛇龙珠虚拟冠层图像,有效地改善数据不平衡问题。同时,相比基于卷积神经网络(Convolutional Neural Networks,CNN)的深度学习模型,基于Transformer的深度学习模型在卷叶病感染程度诊断的问题上更具优势。最佳模型Swin Transformer在增强数据集上达到83.97%的准确率,比在原始数据集上提高3.86%,且高于GoogLeNet、MobileNetV2、NasNet Mobile、ResNet18、ResNet50、CVT和T2TViT等对照模型。而本研究所提的CA-Swin Transformer在增强数据后的测试集上达到86.65%的分类精度,比在原始的测试集上使用Swin Transformer精度提高6.54%。[结论]本研究基于CA-Swin Transformer使用滑动窗口法制作了葡萄园蛇龙珠卷叶病严重程度分布图,为葡萄园卷叶病的防治提供了参考。同时,本研究的方法为无人机遥感监测作物病害提供了一种新的思路和技术手段。[Objective]Wine grapes are severely affected by leafroll disease,which affects their growth,and reduces the quality of the color,taste,and flavor of wine.Timely and accurate diagnosis of leafroll disease severity is crucial for preventing and controlling the dis‐ease,improving the wine grape fruit quality and wine-making potential.Unmanned aerial vehicle(UAV)remote sensing technology provides high-resolution images of wine grape vineyards,which can capture the features of grapevine canopies with different levels of leafroll disease severity.Deep learning networks extract complex and high-level features from UAV remote sensing images and per‐form fine-grained classification of leafroll disease infection severity.However,the diagnosis of leafroll disease severity is challenging due to the imbalanced data distribution of different infection levels and categories in UAV remote sensing images.[Method]A novel method for diagnosing leafroll disease severity was developed at a canopy scale using UAV remote sensing tech‐nology and deep learning.The main challenge of this task was the imbalanced data distribution of different infection levels and catego‐ries in UAV remote sensing images.To address this challenge,a method that combined deep learning fine-grained classification and generative adversarial networks(GANs)was proposed.In the first stage,the GANformer,a Transformer-based GAN model was used,to generate diverse and realistic virtual canopy images of grapevines with different levels of leafroll disease severity.To further ana‐lyze the image generation effect of GANformer.The t-distributed stochastic neighbor embedding(t-SNE)to visualize the learned fea‐tures of real and simulated images.In the second stage,the CA-Swin Transformer,an improved image classification model based on the Swin Transformer and channel attention mechanism was used,to classify the patch images into different classes of leafroll disease infection severity.CA-Swin Transformer could also use a self-attention mechanism to capture the l

关 键 词：无人机遥感 深度学习 生成对抗网络 Swin Transformer 酿酒葡萄卷叶病 数据增强 注意力机制 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程] S436.68[农业科学—农业昆虫与害虫防治]