Disasters, such as building fires, pose substantial hazards to both human existence and ecological integrity. Existing methods sometimes fail to detect fires, especially the semi-occluded/partially occluded fires. Also, traditional disaster response methods rely on human observation, which can be time-consuming and often lack effectiveness. Existing smoke alarms only trigger the alarm when the fire grows to a stage where the majority of damage is done. To address this challenge, our research paper aims to develop a model for the identification of such targets, specifically focusing on semi-occluded fires. By using YOLOv8 to identify the fire, the Double Verification algorithm to assess the color percentages within potential fire regions, and applying Gaussian blur and masking functions the final verification considers color percentages, YOLO confidence, and threshold values to determine fire presence. The results show the effectiveness of our approach for semi-occluded fire detection. Model A utilizing YOLOv8 with a high confidence level (>50%), showed limitations. Model B has lower confidence (>25%), improved detection, but increased false positives. Model C, with lower confidence (>25%) and double verification, outperformed, detecting fires faster with high precision (81%) and fewer false positives. Model C emerges as the most effective choice, demonstrating superior accuracy (0.73) and F1 score (0.81) compared to Model A and B. In conclusion, the integration of a double verification algorithm with YOLOv8 significantly enhances the detection of semi-occluded fires in buildings, as demonstrated by the superior performance of Model C in terms of mean detection time, detection rate, and precision. The utilization of the HSV color space proves to be an effective strategy for identifying chrominance and luminance pixels, offering a promising way of approaching fire detection capabilities in complex environments.