Speaker
Description
More than 90% of the forest territory in Quebec is publicly owned, and as is the forest road network, which is funded through the user-payer principle. There is a publicly available database (AQRéseau+) that contains information on forest roads and their characteristics. Since there are more than 450 000 km of forest roads in the province of Quebec, this database is useful for forest practitioners to support managerial decisions. Accurate forest road inventory is essential for efficient planning of forest operations and for maintaining better control over operational costs. A precise inventory, categorized by road class, enables forest stakeholders to plan their activities appropriately and allocate resources where they are most needed. In contrast, the absence of a reliable inventory often results in the abandonment of certain road sections. Once abandoned, these roads receive little to no maintenance, which accelerates their degradation and increases the risk of environmental impacts such as erosion, sedimentation, and watercourse disturbance.
Currently in Quebec, this data on forest roads is not actively updated beyond the initial input, which limits its usefulness. This project aims to develop a methodology to automate detection and classification of logging roads to have a more reliable knowledge of the forest road network using remotely sensed data.
This project carried out for two regions of the province of Quebec in Canada, the Capitale-Nationale and Gaspésie. It was divided into two phases: first, the detection of existing roads, and then the classification of these roads. First, a literature review was carried out to choose the detection model to be used. This literature review focused mainly on road-detection models using LiDAR technology and their limitations. It also made it possible to identify what kinds of classifications these models can perform. Once the literature review was completed, the modification of an already existing road-detection model was carried out. The original model could detect forest roads and surface water using segmentation and deep learning algorithms. The modified model was applied to forests whose characteristics are representative of the reality of the Capitale-Nationale and Gaspesie regions. Thus, it was tested in both coniferous and deciduous areas. It located roads considered still usable and classified them according to the chosen characteristics. Once the roads were classified, photo interpreters validated the model.
The preliminary results of this project will be presented at the conference.
| Keywords | Characterization; Road; Model; Automatic |
|---|
