Date of Award

12-2025

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

School

Computing Sciences and Computer Engineering

Committee Chair

Dr. Ahmed Sherif

Committee Chair School

Computing Sciences and Computer Engineering

Committee Member 2

Dr. Sarah Lee

Committee Member 2 School

Computing Sciences and Computer Engineering

Committee Member 3

Dr. Dia Ali

Committee Member 3 School

Computing Sciences and Computer Engineering

Committee Member 4

Dr. Chaoyang Zhang

Committee Member 4 School

Computing Sciences and Computer Engineering

Committee Member 5

Dr. Amer Dawoud

Committee Member 5 School

Computing Sciences and Computer Engineering

Abstract

The research work finds a solution to precision agriculture of cotton cultivation using artificial intelligence (AI) models. Two sets of model performance based on the application are selected namely a low resource and a high resource setting. This is because using drone surveys to capture images identifying the classes of stressed and unstressed cotton plantation requires limited model architecture and CPU based computation. Thus, traditional AI models were selected for low resource settings. Again, for high computation intensive models like transfer learning-convolution neural network (CNN) based architectures were grouped into high resource settings. There was another issue of class imbalance in cotton available dataset causing overfitting of model performance. We adopted a target data augmentation strategy to add synthetic data to original dataset, such that each class for classification contributes equally to overcoming the class imbalance. Traditional ML models like Naïve Bayes, K-Nearest Neighbor (KNN), Support Vector Machine (SVM) and Random Forest (RF) were trained with augmentation strategy on cotton disease dataset from Kaggle. Various DL were optimized for better performance like multilayered CNN, VGG16, ResNet50, InceptionV3, Efficient Net, and DenseNet121. Those embodied frameworks were also trained with same dataset with target augmentation to overcome the class imbalance. Out of the models RF and DenseNet121 performed best based on all evaluation metrics and were used for distributed federated learning paradigms with different client data distribution. RF showed the maximum gain in accuracy of 83.69% with data augmentation and enhancement of all other metrics of Precision, Recall, F1 Score and Area Under the Curve (AUC) with targeted augmentation of dataset. Similarly, DenseNet121 showed accuracy of 99.58% under geometric data augmentation. The learning capacity of DenseNet121 enhanced in all other metrics with data augmentation. We found the FL frameworks were scalable and robust and multiclient updates the server model with better performance ensuring data privacy on farmers’ private agricultural field information. The FL based RF shows some susceptibility to client distribution and heterogeneity of client updates to global model and convergence of performance, while the FL based DenseNet121 showed more resilient to such variance of client distribution and global model updates.

ORCID ID

0000-0002-0346-1115

Download

  Contact Author

Share

COinS