Abstract: This paper presents a video smoke detection algorithm based on Adaptive Neuro-fuzzy Inference System(ANFIS). The smoke features are extracted from video sequences, and the subtractive clustering is introduced to confirm the fuzzy rule number. The premise parameters and the consequent parameters are updated by hybrid learning rule. The fuzzy inference rules are obtained. Experimental results show that compared with the traditional BP neural network algorithm and Support Vector Machine(SVM) algorithm, the new algorithm has better performance on Receiver Operating Characteristic(ROC) curve.

Key words: video smoke detection, Adaptive Neuro-fuzzy Inference System(ANFIS), subtractive clustering, smoke feature analysis

摘要： 提出一种基于自适应神经模糊推理系统的视频烟雾检测算法。从视频图像中提取烟雾特征，采用减法聚类确定模糊规则数，建立初始模糊系统。通过神经网络的自学习机制调整前提参数和结论参数，确定模糊推理规则。实验结果表明，与传统BP神经网络算法及支持向量机算法相比，该算法具有较优的ROC曲线特性。

关键词: 视频烟雾检测, 自适应神经模糊推理系统, 减法聚类, 烟雾特征分析

CLC Number: 

• TP18