Final Bachelor Thesis - An Integrated UAV-Based Framework for Precision Weed Mapping in Rangelands: From High-Density Detection to Spatial Cover Analysis
This work presents an integrated UAV-based framework designed for the high-density
spatial mapping of Eryngium horridum—commonly known as ’cardilla’—a critical perennial weed species threatening rangeland productivity across South American grasslands. Rather than treating this as a standard sparse detection task, this thesis tackles the ecological challenge through a density-centric approach, leveraging localized tiling strategies and deep learning architectures to process high-resolution aerial imagery. By conducting a systematic architectural benchmark and optimizing the pipeline for resource-constrained edge environments, this research delivers a scalable, high-precision engineering solution that directly enhances autonomous vegetation monitoring and precision rangeland management.
Computer VisionPyTorchTensorRTEdge ComputingDockerFastAPIGeospatial Mapping 