智慧农业下基于联邦学习的水稻病虫害分类研究

Research on rice pest and disease classification based on federated learning under smart agriculture

在农业发展过程中,农作物的健康问题一直是一个重要的研究课题。面向这一课题探讨智能化农业种植过程中的水稻病虫害分类问题。在智能化农业种植场景下,为了提高设备对病虫害的分类准确性,同时保护各设备的数据隐私,提出使用联邦学习来解决各设备间的数据孤岛问题。实验选取了七个预训练模型来提取特征,使用四个指标(准确率、召回率、损失函数和F1分数)来评估不同模型上的性能。实验结果表明,在独立同分布(IID)和非独立同分布(Non-IID)数据下模型VGG19的准确率分别为99.05%和98.48%,表现出较高的鲁棒性和准确率。通过几种实验和指标对比发现,联邦学习的应用提升了设备4.36%的准确率,图像分类模型的收敛时间受到联邦学习轮数round和每轮联邦学习中训练集的训练epoch数的共同影响,并且模型的稳定性随着参与联邦学习的设备数量增加而提高。

In the ongoing process of agricultural development,the health of crops continues to be a pivotal research area.Address‐ing this issue,this paper endeavors to delve into the classification of rice diseases within the framework of intelligent agricultural planting.Within the context of intelligent agricultural planting,this paper advocates the adoption of federated learning as a means to enhance the accuracy of disease classification equipment while safeguarding the data privacy of individual devices,thereby ad‐dressing the data silo problem among these devices.For the experimental phase,seven pre-trained models are meticulously selected to extract pertinent features,and four evaluation metrics—accuracy,recall,loss function,and F1-score—are employed to assess the performance of these models.The experimental outcomes reveal that the VGG19 model achieved remarkable accuracy levels of 99.05%and 98.48%on Independent and Identically Distributed(IID)and Non-Independent and Identically Distributed(Non-IID)data sets,respectively,showcasing its robustness and precision.Through a series of experiments and comparative analyses of vari‐ous indicators,it is conclusively established that the integration of federated learning has enhanced the accuracy of the equipment by a noteworthy margin of 4.36%.Furthermore,the convergence time of the image classification model is influenced by a combina‐tion of factors,including the number of federated learning rounds and the training epochs per round within the training set.Notably,the stability of the model improves as the number of devices participating in federated learning increases.