Detecting fires is crucial in protecting people, the economy, and the environment by limiting damage. Security personnel have the task of quickly observing fires in open areas, but they might overlook signs of fire or lose track, causing delays and more significant burned areas. Machine learning, which detects fires inside and outside, has gained much attention recently. Experts have created several fire detection systems using computer vision, and CNNs have shown promising results. However, these systems are still not perfect in detecting smoke and fire, especially in open spaces, which are the main reasons for wildfires and can lead to natural disasters. The size, shape, color, and texture of smoke and fire can vary. As a result, even a well-trained automatic system might find it harder to detect and identify fire/smoke than rigid objects with fixed shapes. This study proposes an automatic fire and smoke detection method that builds on the YOLOv8s CNN architecture and adjusts its hyperparameters to create a feasible detector with few false alarms. Experiments done on the D-Fire dataset demonstrate our model’s applicability and high performance compared to the most recent fire and smoke detection systems in the field.