High efficiency and high speed flame detection plays an important role in preventing fire and protecting social security.In this paper,a flame detection method based on Fast-CAANet is proposed to meet the needs of social security applications.Firstly,a CAA module is proposed to strengthen the effective integration of convolution and attention mechanism.Then the main network of CAANet(CAABlock)is constructed to extract the rich characteristics of flame more effectively.The Fast-CAABlock module with smaller parameters and higher accuracy is proposed to enhance the flame feature extraction scheme.Experimental results show that the accuracy of Fast-CAANet is up to 91.42%,and the calculation amount is 3.9 GMac.Compared with other algorithms,the proposed flame detection algorithm has better performance and effect.
deep learningfeature extractionattention mechanismflame detection
万方数据
维普
