Abstract:In the encoder module, the Xception backbone network is being replaced by MobileNetV2, resulting in an effective optimization of network parameters. Adding CBAM to the encoder part to focus on smoke in the complex scenario, which can enrich more details in deep‐level feature maps. The combination of the Dice_Loss function and BCE_Loss (Binary Cross Entropy Loss) function of segmentation is proposed for joint optimization of this network. Fires not only cause devastating consequences for human life and property, but also lead to soil erosion in forests. Therefore, it is necessary to design a novel algorithm that can quickly monitor smoke from fires. Most existing smoke segmentation methods do not consider the segmentation accuracy of algorithms under limited computational resources. To address this research gap, this paper proposes a lightweight smoke segmentation algorithm based on DeepLabV3+ that achieves fast inference speed and high accuracy for different sizes smoke. To reduce the number of parameters, the feature extraction network of the DeeplabV3+ algorithm is replaced by MobileNetV2, which enhances the extraction ability of the algorithm in segment smoke images. Then, the Convolutional Block Attention Module (CBAM) is added to the encoder part to enhance the perception of the algorithm for small smoke and effectively alleviates smoke mis‐segmentation. Furthermore, a newly designed loss function is used in the network. The experimental results show that the proposed method has improved by 1.27% in Smoke IoU and 1.21% in mPA compared with other methods. The weight size has been reduced to 10.76% of the original DeepLabV3+, and the inference time is only 33.71ms. Therefore, it is a more suitable early fire detection algorithm for resource‐constrained environments.

Candidate Smoke Region Segmentation of Fire Video Based on Rough Set Theory

A Slight Smoke Perceptual Network.

Rough-fuzzy set approach for color and texture based image segmentation

A smoke segmentation algorithm based on improved intelligent seeded region growing

Level Set Based Flame Segmentation

Deep Smoke Segmentation

Unsupervised Segmentation of Fire and Smoke from Infra-Red Videos

A Modified Method of Video-based Smoke Detection for Transportation Hub Complex

An Algorithm for Smoke RoF Detection Based on Surveillance Video.

Video Smoke Detection Based on Deep Saliency Network

Smoke Object Segmentation and the Dynamic Growth Feature Model for Video-Based Smoke Detection Systems

A lightweight network for real-time smoke semantic segmentation based on dual paths

FoSp: Focus and Separation Network for Early Smoke Segmentation

Dynamic smoke detection by eliminating static targets in video

Upper Body Human Detection and Segmentation in Low Contrast Video

Multi-Scale Flame Situation Detection Based on Pixel-Level Segmentation of Visual Images

Wavelet based smoke detection method with RGB Contrast-image and shape constrain

A lightweight multiscale smoke segmentation algorithm based on improved DeepLabV3+

Target segmentation of industrial smoke image based on LBP Silhouettes coefficient variant (LBPSCV) algorithm

Smoke root detection from video sequences based on multi-feature fusion

Smoke detection method based on red, green and blue (RGB) contrast image and target shape