Improved lightweight flame smoke detection algorithm for YOLOv8n

To address the problems of untimely early fire detection and early warning and low precision, target detection of early small flames with leakage, and large number of model parameters, a lightweight flame smoke detection algorithm is proposed to improve the You Only Look Once version 8 (YOLOv8) method. In terms of network structure, first, the bottleneck in the Context module to fuse features (C2f) module of YOLOv8 is replaced by FasterNetBlock in FasterNet in the backbone network to make the network lightweight and reduce the number of parameters of the network model; second, the Convolutional Block Attention Module (CBAM) is introduced so that the network can improve the ability to extract features from the channel and the space to avoid the problems caused by the lightweighting of the network model. To enhance the detection of small targets, a new small target detection header with an output size of 160 × 160 pixels is added, which is generated by the first C2f module in the backbone network. Finally, the Enhanced Intersection over Union (EIOU) loss function is introduced as the target frame regression loss function to improve the imbalance between positive and negative samples in the dataset. In terms of the dataset, fire scenes with different backgrounds are selected, including the early, middle and late stages of the fire, and 15% of sunset, lighting and other interference images are added for a total of 6600 images. For the presentation of the results, a visualization interface was designed based on Streamlit. Compared with the previous model, the improved YOLOv8n improves the model accuracy by 8.4%, while the model size increases by 0.1%, which effectively achieves EIOU a balance between model weight and performance.