Computer Vision
[Open soon]
Work description
This research fellowship aims to develop computer vision methods for the automatic analysis of images of pastures and soils, collected via satellite, drone, rover and/or devices attached to animals, within the context of the DFence project. The work will focus on creating image processing and annotation pipelines, developing classification, detection and segmentation models, extracting georeferenced indicators on the state of vegetation cover and soil, and integrating these indicators into the Decision Support System to aid the dynamic definition of virtual fences and the sustainable optimisation of grazing. A particularly important aspect is the implementation of self-service approaches, which allow producers to integrate their own image sources (e.g. drones, rovers, cameras) and configure the pipelines to use them, including their participation in the process (e.g. image annotation), so as to enable each implementation to be adapted to specific use cases without requiring technical knowledge. Specifically, the main activities to be carried out by the student are: • Define a visual taxonomy of the grazing and soil conditions relevant to DFence: vegetation cover, bare soil, waterlogged soil, degraded areas, overgrazed areas and areas suitable for grazing. • Build a pipeline for the collection, pre-processing, annotation and management of images from different sources: satellite, drone, rover and/or cameras attached to animals’ devices. • Develop image classification, detection and/or segmentation models to automatically estimate indicators of pasture and soil condition. • Evaluate Deep Learning models suitable for the problem, including CNNs, semantic segmentation architectures and object detection models. Relevant families of techniques/architectures include YOLO, SSD, CNNs, ResNet, VGGNet, multispectral/hyperspectral analysis and tracking. • Integrate the model outputs in a format compatible with the data pipeline and the Integrated Decision Support System, producing georeferenced indicators by grazing area. • Validate the models using field data, environmental sensors and agro-environmental indicators, measuring performance using metrics such as accuracy, F1-score, IoU, mean absolute error, correlation with field measurements and robustness under different lighting conditions, seasons and terrain types.
Academic Qualifications
- A degree in computer engineering, information systems or a related field;
Minimum profile required
A bachelor's degree with an average grade of over 12.
Preference factors
Fluency in Portuguese. - Experience in computer vision techniques, specifically in self-service approaches focused on technology adoption issues. - Preference will be given to candidates with a master’s degree; candidates with a bachelor’s degree will only be considered if no master’s degree holders apply, or if those who do apply do not have the required background and/or the experience mentioned above.
Application Period
Since 14 May 2026 to 27 May 2026
[Open soon]
Centre
Industrial & Systems Engineering and Management