Vegetation Detection and Classification for Power Line Monitoring

• Autor(a) principal: Faustino, Pedro Miguel Freire
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/146013
• topic: Vegetation Management; Power Line Mapping; Remote Sensing; UAV; Point Cloud Classification; Point Cloud Filtering; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Electrical network maintenance inspections must be regularly executed, to provide a continuous distribution of electricity. In forested countries, the electrical network is mostly located within the forest. For this reason, during these inspections, it is also necessary to assure that vegetation growing close to the power line does not potentially endanger it, provoking forest fires or power outages. Several remote sensing techniques have been studied in the last years to replace the labor-intensive and costly traditional approaches, be it field based or airborne surveillance. Besides the previously mentioned disadvantages, these approaches are also prone to error, since they are dependent of a human operator’s interpretation. In recent years, Unmanned Aerial Vehicle (UAV) platform applicability for this purpose has been under debate, due to its flexibility and potential for customisation, as well as the fact it can fly close to the power lines. The present study proposes a vegetation management and power line monitoring method, using a UAV platform. This method starts with the collection of point cloud data in a forest environment composed of power line structures and vegetation growing close to it. Following this process, multiple steps are taken, including: detection of objects in the working environment; classification of said objects into their respective class labels using a feature-based classifier, either vegetation or power line structures; optimisation of the classification results using point cloud filtering or segmentation algorithms. The method is tested using both synthetic and real data of forested areas containing power line structures. The Overall Accuracy of the classification process is about 87% and 97-99% for synthetic and real data, respectively. After the optimisation process, these values were refined to 92% for synthetic data and nearly 100% for real data. A detailed comparison and discussion of results is presented, providing the most important evaluation metrics and a visual representations of the attained results.