Revealing the population of Western gorillas on salines in the north of the Republic of the Congo using artificial inteligence

Lina Maria Castellanos Vela, BSc

Diplomová práce

Abstract:

This Master's thesis investigates the application of machine learning-based annotations for monitoring Western Lowland Gorillas (Gorilla gorilla gorilla) across Dibo Bai (2.148852N, 14.629689E), in the Messok-Dja area, Republic of Congo. Using camera trap data (9,087 images), the study compares the performance and validity of the YOLO v11 object detection model in making human annotations for gorilla detection (bodies and faces). Furthermore, it analyzes the spatiotemporal distribution patterns, habitat preferences, diel activity using model outputs and manual annotations to establish accuracy, group compositions, and potential signs of yaws disease. Relative Abundance Index (RAI) analysis shows that Dibo Bai gorillas seem to prefer forest edges with RAI= 6.95 and a bimodal diel activity pattern (late morning and afternoon). According to the Wald test results, gorillas appear to be more active later in the day during the dry season (Act= 3.13) compared to the wet season (Act = 2.99). Gorilla body detections by the YOLO model shows a high performance with precision (0.941) and recall (0.926) meaning that there were just a few false positives, and that the model detects almost all bodies of gorillas in the images. In contrast the evaluation of the performance of the model by comparing with the manual annotations shows a RMSE= 0.902 for body predictions.Low prevalence of yaws disease with no statistically significant correlation with social structure. Future research would have to improve the YOLO v11 model's accuracy and reliability, design gorilla tracking for individuals, and study the ecological, environmental, and social factors influencing gorillas behavior and long-term health within the study area.

Keywords

Western Lowland Gorilla, Camera Traps, Spatiotemporal Distribution, YOLO v11, Machine Learning, Messok-Dja, Wildlife Monitoring, Yaws Disease

Jazyk práce: angličtina

Datum vytvoření / odevzdání či podání práce: 5. 4. 2025

Obhajoba závěrečné práce

• Obhajoba proběhla 19. 5. 2025
• Vedoucí: Mgr. Tomáš Jůnek, Ph.D.
• Oponent: Matyáš Adam, externi

Česká zemědělská univerzita v Praze

Fakulta životního prostředí

Magisterský studijní program:
Nature Conservation