Indoor Video Flame Detection Based on Lightweight Convolutional Neural Network
- PDF / 2,032,364 Bytes
- 14 Pages / 612 x 792 pts (letter) Page_size
- 16 Downloads / 267 Views
Indoor Video Flame Detection Based on Lightweight Convolutional Neural Network Zhikai Yanga,*, Leping Bua,**, Teng Wanga,***, Peng Yuana,****, and Ouyang Jinenga,***** a
College of Electrical Engineering, Naval University of Engineering, Wuhan, 430033 China * e-mail: [email protected] ** e-mail: [email protected] *** e-mail: [email protected] **** e-mail: [email protected] ***** e-mail: [email protected]
Abstract—At present, all CNN-based fire detection algorithms identify fire by means of a single frame image, all of which demonstrate low accuracy under strong interferences or complex backgrounds such as flickering light or backgrounds with high level of brightness. To increase the accuracy of fire detection, this paper presents a neural network model which combines lightweight CNN with SRU. In this algorithm, the scene content is extracted by CNN and the dynamic characteristics of the flames are extracted from sequential frames. In this paper, Resnet18+SRU (V1-type) and Mobilenets+SRU (V2-type) are proposed. Based on the characteristics of flames at a fixed position within a short period of time, a 3D convolutional layer is added between the Mobilenets and the SRU in the V2-type model, resulting in the V3-type model. Based on a cross validation set containing multiple types of interference in an indoor environment, experiments were conducted to compare the three models proposed in this paper with other models. The experiment results showed that the accuracy of the method proposed in this paper is above 96%, about 25% higher than the accuracy of CNN-based fire alarm via single-frame image, and that the V3-type models with 3D convolutional layer has the highest accuracy and best overall performance. Keywords: flame alarm, convolutional neural network, simple recurrent unit, 3D convolutional layer DOI: 10.1134/S1054661820030293
1. INTRODUCTION Traditional fire detection algorithms mostly use manually selected features as vectors for distinguishing fire from interference. Chen et al. chose the three grayscale levels of the RGB channel of images as variables and used a simple and quick evaluation function to detect the flame in an image. It has remarkable effects on forest fires with simple green background [1]. Wang et al. improved the algorithm by adding a dispersion function of the blue channel of the flame region to the evaluation function. It is effective for indoor flames [2]. However, both algorithms used few features and the evaluation function threshold was manually set according to the specific data set. In some scenarios where there are many changes and interferences, it is necessary to adjust the parameters repeatedly resulting in low accuracy. With the continuous improvement of convolutional neural networks (CNN) in image categorization and the decrease in computational complexity, CNN is gradually applied to process fire detection. Two highly computational CNN models, VGG16 and Res-
Received January 13, 2019; revised January 13, 2019; accepted January 28, 2020
net50 are used to identify whet
Data Loading...
