摘要
针对传统火灾探测器在火灾检测方面的不足,传统的图像识别技术对于火灾预测精度不高的问题,采用基于改进的RFBNet算法进行火灾识别。本算法在原来的RFBNet算法的框架上进行改动,采用自适应特征提取优化特征提取操作、反卷积增强融合和增强型软性非极大值抑制,通过提高网络的特征提取能力、加强层级联系和减少重叠目标的漏检率来提高小目标物体的识别精度低和识别准确率低的问题。实验表明,该方法可以有效识别出火灾,识别精度达到93.5%,优于RFBNet算法,速度上也满足实时性要求。
Aiming at the shortcomings of traditional fire detectors in fire detection, traditional image recognition technology has low fire prediction accuracy, so the improved RFBNet algorithm is used to recognize fire. In this algorithm,the framework of the original RFBNet algorithm is modified, and adaptive feature extraction is adopted to optimize feature extraction operation, deconvolution enhanced fusion and enhanced soft non-maximum suppression are adopted. By improving the feature extraction ability of the network, strengthening hierarchical connection and reducing the missed detection rate of overlapping targets, the problems of low recognition accuracy and low recognition accuracy of small target objects are solved.The experiment result shows that this method can effectively identify the fire, and the recognition accuracy reaches 93.5%,which is superior to RFBNet algorithm, and the speed also meets the real-time requirements.
出处
《科技创新与应用》
2022年第27期14-17,共4页
Technology Innovation and Application
