Multi-Scenario Fire Detection Algorithm Based on Improved YOLOv7

To address the issue of reduced detection efficiency in traditional fire alarm systems owing to environmental and spatial limitations during fire incidents, this study investigated an improved YOLOv7 algorithm specifically designed for multi-environment fire detection. A dynamic head was integrated into the original algorithm to enhance the detection accuracy, and a global attention mechanism and slim neck model structure were introduced to improve the extraction of fire-related features and reduce the computational complexity. Additionally, the Wise-IoUv3 loss function was employed instead of the traditional IoU loss function in the model, and the training weights were adjusted based on the image quality to optimize the model efficiency. In an experiment using a multi-environment fire dataset, the improved YOLOv7 algorithm achieved an mAP@0.5 of 71.1%, which was a 5.3 percent improvement over the result with the original algorithm. This demonstrated a high level of detection accuracy and effectively met the requirements for precise fire detection in complex environments.