Abstract:

The traditional fire detection algorithm relies on artificial feature engineering, which is subjective and blind. It has the problems of poor generalization ability and low detection accuracy. In this paper, a fire image detection algorithm based on improved Mask R-CNN is proposed. On the basis of Mask R-CNN, the algorithm introduces a bottom-up feature fusion in the feature pyramid, and improves the loss function to make the frame location more accurate. Experiments on the self-built test dataset show that the accuracy of the improved algorithm is higher than that of the original algorithm, and the detection accuracy is increased by more than 5%.

Key words: fire detection, feature engineering, residual network, Feature Pyramid Network（FPN）

摘要：

传统火焰检测算法依赖于人工特征工程，具有主观性和盲目性，存在泛化能力差，检测准确率不高等问题。提出一种基于改进Mask R-CNN的火焰图像识别算法，算法在Mask R-CNN的基础上，在特征金字塔引入一条自下向上的特征融合，同时改进了损失函数，使边框定位更准确。在自建的测试数据集上实验表明，改进后算法准确率相对于原先算法识别定位精度更高，检测准确率提升超过5%。

关键词: 火灾检测, 特征工程, 残差网络, 特征金字塔网络