SMART IMAGING FOR AGRICULTURE DATA WITH ROAD DETECTION INTEGRATION
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
UNIVERSITI MALAYSIA SARAWAK
Abstract
Description
This paper offers a comparative assessment of three deep learning models (U-Net, DeepLabV3, and YOLOv8n-seg) for road segmentation in agricultural settings utilising advanced imagery techniques. The aim of this research is to provide an AI-driven solution that identifies unstructured farm roads in aerial pictures using convolutional neural networks and semantic segmentation techniques. A dataset was created via preprocessing and augmentation, facilitating efficient model training and assessment. Each model was executed and evaluated using metrics (IoU, Dice), threshold sensitivity, and generalisation on novel data. U-Net attained superior segmentation accuracy with continuous and intricate masks, but DeepLabV3 demonstrated consistent performance across diverse data volumes. YOLOv8n-seg, while efficient in speed, produced fragmented predictions inadequate for fine-grained segmentation. The results underscore the trade-offs between accuracy, robustness, and efficiency among the models. This study advances the creation of accurate and scalable road identification systems for smart agriculture, including suggestions for future enhancements in boundary refining, adaptive thresholding, and multi-modal data fusion.