The identification of infesting plants in monoculture fields is a critical task for enhancing crop yields and promoting sustainable agriculture. This thesis investigates the use of Convolutional Neural Networks (CNNs) for the automated detection and classification of infesting plants in agricultural settings using high-resolution imagery obtained from Unmanned Aerial Vehicles (UAVs). Various CNN architectures, including ResNet50, VGG16, InceptionV3, Inception- ResNetV2, and DenseNet201, were evaluated to determine their effectiveness in distinguishing between crops and infesting plants. The study demonstrates that DenseNet201 outperforms other models, achieving the highest accuracy due to its dense connectivity pattern, which ensures efficient gradient flow and feature reuse. This architecture’s ability to learn complex and abstract features makes it particularly adept at handling the variability and complexity present in UAV imagery. The thesis addresses the challenges of dataset preparation, including data augmentation and class imbalance, and highlights the potential of integrating UAVs and CNNs for real-time agricultural monitoring. The impact of false positives (FP) and false negatives (FN) is also carefully examined, given their significant implications in crop monitoring, by introducing methodologies to control the values of precision and recall. The results indicate that this approach can significantly reduce labor costs and enhance crop management efficiency. Future research directions include improving model generalization through diverse datasets, exploring advanced CNN architectures, developing real-time processing methods, and integrating automated intervention systems. By addressing these areas, the research aims to contribute to more effective and sustainable agricultural practices.
Identification of infesting plants in monoculture fields through CNNs from UAV imagery
VO, FRANCESCO
2023/2024
Abstract
The identification of infesting plants in monoculture fields is a critical task for enhancing crop yields and promoting sustainable agriculture. This thesis investigates the use of Convolutional Neural Networks (CNNs) for the automated detection and classification of infesting plants in agricultural settings using high-resolution imagery obtained from Unmanned Aerial Vehicles (UAVs). Various CNN architectures, including ResNet50, VGG16, InceptionV3, Inception- ResNetV2, and DenseNet201, were evaluated to determine their effectiveness in distinguishing between crops and infesting plants. The study demonstrates that DenseNet201 outperforms other models, achieving the highest accuracy due to its dense connectivity pattern, which ensures efficient gradient flow and feature reuse. This architecture’s ability to learn complex and abstract features makes it particularly adept at handling the variability and complexity present in UAV imagery. The thesis addresses the challenges of dataset preparation, including data augmentation and class imbalance, and highlights the potential of integrating UAVs and CNNs for real-time agricultural monitoring. The impact of false positives (FP) and false negatives (FN) is also carefully examined, given their significant implications in crop monitoring, by introducing methodologies to control the values of precision and recall. The results indicate that this approach can significantly reduce labor costs and enhance crop management efficiency. Future research directions include improving model generalization through diverse datasets, exploring advanced CNN architectures, developing real-time processing methods, and integrating automated intervention systems. By addressing these areas, the research aims to contribute to more effective and sustainable agricultural practices.File | Dimensione | Formato | |
---|---|---|---|
vo_tesi_definitiva.pdf
accesso aperto
Dimensione
24.68 MB
Formato
Adobe PDF
|
24.68 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/71039