In principle, glacial lakes originating from accelerated retreat under climate change are growing rapidly and they are increasingly threatened through deep segregation reactivity which is vis-a-vis Glacial Lake Outburst Floods (GLOF). Therefore, alerting people in advance and keeping a watch on these water bodies is very critical for disaster risk reduction and environmental management. Classic methods involving manual inspection and classical remote sensing are slow, punitive and are not suitable for immediate intervention. In this paper, we propose a novel deep learning approach based on the YOLOv8 object detection and segmentation algorithm which will be used for automatic extraction of glacier lakes from high-resolution satellite imagery. Integrating the anchor-free detection module, YOLOv8 features a segmentation head, which makes it a highly efficient and effective solution to simultaneously tackle the delineation and localization of glacial lakes with highly efficient results. The model was trained and fine-tuned on a curated dataset of images of glacier lakes and included data augmentation techniques to increase generalization. Through experimental results we find that the proposed framework provides competitive results in terms of precision, recall, mean Average Precision (mAP) and maintains real-time inference speeds that are acceptable for the application of pipeline health monitoring. Also, the trained model is embedded in an alarm system which displays audible warnings in the case of detection of the glacier lakes to simulate the warning as an early warning. It demonstrates many ways in which modern computer vision techniques can respond in tandem with environmental monitoring infrastructure in an effort to address climate induced hazards. Future effort shall be on scale up of the dataset, multi-spectral imagery incorporation and predictive hydrological modelling for proactive GLOF risk assessment