Forest fires have emerged as a significant threat to the environment, wildlife, and human lives, necessitating the development of effective early detection systems for firefighting and mitigation efforts. In this study, we introduce TeutongNet, a modified ResNet50V2 model designed to detect forest fires accurately. The model is trained on a curated dataset and evaluated using various metrics. Results show that TeutongNet achieves high accuracy (98.68%) with low false positive and false negative rates. The model's performance is further supported by the ROC curve analysis, which indicates a high degree of accuracy in classifying fire and non-fire images. TeutongNet demonstrates its effectiveness in reliable forest fire detection, providing valuable insights for improved fire management strategies.