Fig 1.
Overall network architecture of FCMI-YOLO.
Fig 2.
(a) FasterNet. (b) S-FasterNet. (c) FasterNext.
Fig 3.
The principle of Partial Convolution.
Fig 4.
Comparison curve of ReLU and SiLU activation functions.
Table 1.
Parameters of the FasterNext and C3.
Fig 5.
The principle of MLCA mechanism.
Fig 6.
The structure of MLCA mechanism.
Fig 7.
Schematic diagram of Inner-IoU.
Fig 8.
Distribution of the dataset.
Table 2.
Parameters of the dataset.
Table 3.
Model train environment.
Table 4.
Primary training parameters for the model.
Table 5.
Performance of fire detector based on different model versions of YOLOv5.
Table 6.
Performance comparison of YOLOv5s with FasterNext module replacement.
Fig 9.
Comparison of detection results of different methods in FasterNext.
(a) YOLOv5s. (b) YOLOv5s + FasterNext (Backbone). (c) YOLOv5s + FasterNext(Neck). (d) YOLOv5s + FasterNext(Backbone + Neck).
Fig 10.
Comparison of mAP@0.5 for different ratios.
Table 7.
Performance comparison of different loss functions.
Table 8.
Ablation experiments results of YOLOv5s.
Fig 11.
Detection results of YOLOv5s and FCMI-YOLO under different exposure levels.
(a) Fire in the close interior. (b) Remote outdoor fire. (c) Fire in the Night.
Fig 12.
Comparison of mAP@0.5 and Recall for mainstream algorithms.
Table 9.
Performance comparison of mainstream algorithms.
Fig 13.
Detection results of FCMI-YOLO, YOLOv6s, YOLOv9s, and YOLOv11s.
Fig 14.
System diagram.
Table 10.
NPU utilization and FPS Under different processing methods.
Fig 15.
mAP performance of FCMI-YOLO on the Orange Pi 5 Plus.
Fig 16.
Detection results of FCMI-YOLO and YOLOv5s at 30m and 75m on the OrangePi 5 Plus.