A smoke segmentation algorithm based on improved intelligent seeded region growing

Abstract

Summary Image segmentation method based on region growing has the advantages of simple segmentation method and complete segmentation target. Because the color discrimination and gray gradient of smoke are not obvious, the traditional region growing segmentation method is difficult to separate it from the image, resulting in an unsatisfactory segmentation effect. To solve this problem, this paper partially improved the region growing method and proposed a new smoke segmentation algorithm based on the improved intelligent seeded region growing (IISRG) method. First, smoke images obtained from experimental videos were converted from the RGB color space to the HSV color space, and image binarization was achieved using background subtraction with an adaptive threshold in the V channel. Then, a pixel in the binary image was selected intelligently as the seed point, which was used for the regional growth. The final smoke segmentation images were obtained by the morphological processing of region growing images. Experimental smoke segmentation results show that the proposed algorithm has a higher overlap rate and a lower overflow rate, and performs a better smoke segmentation effect compared with the other two approaches. In addition, this algorithm can also effectively solve the problems of under‐segmentation and over‐segmentation of smoke images.

An efficient fire detection algorithm based on multi‐scale convolutional neural network

“…Furthermore, BoWFire‐Dataset 28 is tested and contrasted with several networks (LeNet, 24 Chino et al, 29 Rudz et al, 30 Rossi et al, 31 Celik et al, 8 Chen et al 6 ) to verify the effectiveness of the performance of our algorithm. After training in a fixed size (400 × 400) on our dataset according to Section 2.4, the original network LeNet is also tested.…”

“…However, both are always limited by the reliability of the sensors, laying position, setting parameters, etc., and show the defects of low accuracy and high false alarm rate, 5 especially in ample space or irregular areas. On the contrary, video‐based fire detection (VFD) technology 6,7 is very attractive for military, social security, and commerce for the advantages of high visualization, quick response, real‐time self‐diagnosis, large protection area, and the ability to store and playback videos.…”

Full-Scale Fire Smoke Root Detection Based on Connected Particles

“…The smoke root detection method proposed in this paper was compared with two similar methods, including a forest fire smoke detection system based on the visual smoke root, which is a diffusion model proposed by Gao et al (Method 1) [16], and a smoke segmentation algorithm based on improved intelligent seeded region growing (Method 2), proposed by Zhao et al [27]. It is worth noting that, since the purpose of Zhao's method is to detect smoke, not smoke roots, in order to compare the proposed method with Zhao's method, we combined the developed smoke root node method in this paper with Zhao's smoke detection so as to obtain the final results of Zhao's method.…”

An Image-Based Fire Monitoring Algorithm Resistant to Fire-like Objects

“…Compared with traditional image-based fire detection algorithms, deep learning technologies present a novel approach to visual-based fire detection. They have demonstrated outstanding performance in automatic feature extraction, coupled with high accuracy, enhanced speed, reliable operation, and cost-effectiveness [22][23][24][25][26]. Frizzi et al classified images as fire, smoke, and no fire using a six-layer convolutional neural network (CNN) [27].…”