Accurate and real-time fire detection is crucial for industrial production and daily life. However, the variable form of fire and the significant differences in visual characteristics across its different stages pose great challenges to precise fire prevention and control. To address this issue, a multi-scale fire target detection algorithm using YOLO-fire was proposed by improving the YOLOv8 model. This model introduced new layer structures and attention mechanism, replaced new feature fusion modules and loss functions. By introducing a small-target detection P2 layer, the model’s ability to detect early-stage fires is improved. The coordinate attention mechanism is integrated into the layer structures of multi-scale target detection, enhancing the capture of target location information and channel relationships, thereby focusing more on the target regions. The Neck network structure was optimized by adopting a BiFPN_F strategy for different feature layers, which strengthened the cross-scale representation of fire features and controlled the parameter count of the designed model. The WIoU loss function was employed to optimize the regression process, improving fire source localization accuracy in complex scenarios, enhancing model robustness, and increasing detection precision. Experimental results on fire datasets demonstrated that YOLO-fire could effectively detect multi-scale fire targets in various scenarios. Compared to the baseline model (YOLOv8n), YOLO-fire achieves improvements of 1.37% in accuracy, 1.25% in mAP50-95, and 0.35% in F1-score, while reducing parameters by 3.79%. Furthermore, compared to current mainstream target detection algorithms, YOLO-Fire achieved optimal detection performance while reducing network parameters and computational complexity. This research provided effective technical support for fire safety prevention and control in related fields.

In order to solve the problems of high false positive and false negative rates of external fire monitoring in complex mine environments, a monitoring algorithm using infrared visual feature fusion was proposed. Firstly, the Local Contrast Measure (LCM) model for infrared small target detection was improved to enhance the saliency of early-stage fire targets, thereby segmenting out suspected fire areas. Then, by analyzing the visual features of exogenous fires and major interfering heat sources in thermal infrared image sequences under different surveillance scenarios, the salient features of fires with strong anti-interference ability were preferred. Next, fire salient feature extraction methods and similarity estimation strategies were optimized to obtain the main visual features of suspected fire areas in the thermal infrared image sequences and construct a fire feature vector. Finally, by establishing a feature vector set and constructing a mine exogenous fire detection model using Support Vector Machine (SVM), the proposed algorithm was experimentally validated. The results show that the proposed algorithm realizes exogenous fire monitoring in different scenarios, as well as in remote and early stages, with accuracy and detection rates of 96.93 % and 96.24 %, respectively, and a false detection rate of 2.56 %. Compared to the described comparison algorithms, the proposed method has better improvements in the accuracy, false alarm rate, and leakage alarm rate of fire monitoring.

In order to overcome the problems of poor real-time performance, high false alarm rate and underreport alarm rate of mine external fire monitoring, a method of fire monitoring using the fusion of visible visual features is proposed.Firstly, the visual features corresponding to the video images of fire sources in different monitoring environments are analyzed, and the extraction methods of fire source texture, sharp corners, similarity coefficient and flicker frequency are designed.Then, an improved seed region growth algorithm is used to segment the suspected fire area, and different feature extraction methods are used to calculate the dynamic and static characteristics of the suspected fire area.Secondly, the extracted dynamic and static features are used to construct fire feature vectors.Finally, a fire monitoring model using BP neural network is constructed, and monitoring model is verified.The results show that the proposed fire monitoring method can effectively detect mine external fire in different scenes and distances.The correct rate and detection rate are 98.60% and 99.06%, respectively, the false detection rate is as low as 2.00%.It has strong anti-interference ability and robustness.