Abstract
Deep learning applications are being adopted in agricultural image analysis that include challenges of data privacy and limited institutional data and heterogeneity of different types of architectures. However, Federated Learning is a model that allows collaborative training on data that does not have to be shared among parties. Therefore, Federated Learning is an effective method of collaborative training; however, its comparative effectiveness as compared to individual (local) training on diverse architectures has never been examined in an agricultural context. The objective of this study was to examine Federated Learning for the purpose of crop disease classification on extreme non-IID distributed data sets across three architectures (ResNet-18, EfficientViT, and ConvNeXt), utilizing two unique data sets that were obtained from the laboratory (PlantVillage) and field (PlantWild) environments. The Federated Averaging algorithm was used to perform 30 rounds of federated averaging and utilized 2 local epochs per round. Federated Learning demonstrated comparable performance as local training. Specifically, ResNet-18 increased from 90.5% to 92.3% accuracy (+1.8%), EfficientViT increased from 91.2% to 93.1% (+1.9%) indicating that the collective learning process enabled better generalization of the models because of collaboration. In contrast, ConvNeXt decreased from 95.7% to 94.9% (-0.8%) indicating that high-capacity architectures may experience challenges when optimizing in a federated environment. Additionally, Federated Learning achieved an average accuracy of 93.4% as compared to 92.5% using local training and had significantly higher ROC-AUC scores than local training (ResNet: 0.967 vs 0.958; EfficientViT: 0.971 vs 0.962). Findings indicate that Federated Learning can be effectively implemented to support low-to-moderate capacity architectures to enable collaborative training; however, it will be important to consider the potential limitations of high-capacity architectures when implementing Federated Learning.
Advisor
Rushit Dave
Committee Member
Rajeev Bukralia
Committee Member
Mansi Bhavsar
Date of Degree
2026
Language
english
Document Type
Thesis
Degree
Master of Science (MS)
Program of Study
Data Science
Department
Computer Information Science
College
Science, Engineering and Technology
Recommended Citation
Janga, R. (2026). Impact of cross-client heterogeneity in federated learning for real-world plant disease classification [Master’s thesis, Minnesota State University, Mankato]. Cornerstone: A Collection of Scholarly and Creative Works for Minnesota State University, Mankato. https://cornerstone.lib.mnsu.edu/etds/1595/
Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.
Included in
Agricultural Science Commons, Artificial Intelligence and Robotics Commons, Data Science Commons
Rights Statement
