复杂环境下小尺度烟火目标检测研究

针对复杂环境下起火点目标尺寸较小、起火点特征易与实际场景混淆导致烟火检测效率及准确率低等问题，提出了一种基于改进YOLOv5的小尺度烟火目标检测方法.首先，在原始YOLOv5模型输出的第3个检测层上增加第4个检测层，以此获取更大的特征图对小目标进行检测，加强网络模型的特征提取能力.其次，为解决目标在被遮挡的场景中容易出现漏检的问题，将原网络中用于计算目标框损失函数的GIoU_Loss替换成DIoU_Loss.最后，利用TensorRT对模型进行压缩和加速优化，并将其部署到Jetson TX2开发板上进行加速推理实验，通过复制增强方法扩充实际烟火场景数据.大量实验结果表明，本文所提方法用于复杂环境下的小尺度烟火目标检测不仅检测速度快而且精度高，适于推广应用.

Small scale smoke & fire target detection in complex environment

To address the low efficiency and accuracy of smoke & fire detection due to the small size of target and the confusion of fire feature with actual scene in complex environment, a small scale smoke & fire target detection method based on improved YOLOv5 is proposed.First, a fourth detection layer is added to the third detection layer output in the original YOLOv5 model, so as to obtain a larger feature map for small target detection and strengthen the feature extraction capability of the network model.Second, to solve the easy missing detection of target in shielded scene, DIoU_Loss is used to replace the GIoU_Loss in calculating the regression loss function of the target frame.Finally, TensorRT is used to compress and accelerate the optimization of the model, and then deployed to the Jetson TX2 development board for accelerated inference experiments.In addition, more smoke & fire scene data are constructed by replication enhancement.Experimental results show that the proposed method has fast convergence speed and high accuracy for small scale smoke & fire detection, possessing the prospect for popularization and application.