Détection de fuites d’eau par drone à l’aide de l’imagerie thermique et de l’analyse du réseau de drainage

Abstract

Environmental risk management in the mining sector is a major issue to limit the risks of contamination of natural environments and the consequences on human health. Tailings ponds, often open-air, concentrate a large quantity of toxic metals representing a potential source of contamination. Due to their high mobility following events such as precipitation, water infiltration, or freeze-thaw cycles, increased monitoring of these mining sites is therefore essential. This monitoring involves the detection and prevention of possible leaks or wetlands that would cause the dispersion of trace metals and lead to the degradation of surrounding ecosystems. In this context, technological advances in drones and imaging sensors offer new opportunities. The study presented was conducted at the Quémont 2 mining site of the Horne smelter, located in Rouyn-Noranda, Quebec. Some of the site's aging retention structures may experience water losses at their base. Locating leaks is often very difficult without technological means due to the site's large surface area and the presence of vegetation, shaded areas, and rocks of various sizes. To address this, the method presented proposes to simultaneously exploit thermal and visible images captured by a drone. Infrared (IR) imaging makes it possible to detect temperature contrasts, probable indicators of the presence of water, while high-definition visible images are used to generate secondary rasters (vNDVI, shading, slopes) that help filter false positives included in thermal images. For example, vNDVI (visible NDVI) allows the isolation of vegetated surfaces that may have a thermal signature similar to that of water. Slopes and shading help identify artifacts caused by topography or shadows. Data thresholding using machine learning algorithms is also employed, improving processing accuracy and the autonomy of the approach while reducing analysis time. Finally, this imaging preprocessing enables the tracing of a drainage network to distinguish possible groundwater runoff from surface runoff. The results obtained show that this integrated approach effectively isolates resurgences and highlights observed seepages, making it possible to identify potential sources of exfiltrations, thereby facilitating the planning of human operations and decision-making. The proposed method is adaptable to various environmental monitoring applications. Its current limitations are mainly related to the performance of the thermal sensor and the meteorological conditions at the time of image acquisition.

Date	26 Mar 2025
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Michel Baraër (Supervisor) & Eric Rosa (Co-supervisor)