基于U2-Net和VGG16的天然气泄露检测

目前在深度学习领域很少以天然气泄露图像为数据进行研究，本文使用甲烷红外图像训练的卷积神经网络(VGG16)来实现泄露检测。另外，针对泄露的甲烷气体与背景图像存在相似性的问题，使用U²-Net图像分割网络代替背景建模方法来提取泄露气体区域。通过迁移VGG16网络模型结构和卷积层参数，在卷积层和激励层之间加入BN层以提高训练速度，将最后一层池化层替换为基于最大池化算法的动态自适应池化方法以提高检测精度。将改进的VGG16神经网络对分割的红外图像进行训练并与其他卷积神经网络进行对比，使用准确率，精准率，召回率和F1-score来对模型进行综合评价，其表现效果最好。与现有的检测方法进行对比，所提出的检测方法准确率更高。该检测方法能够实现高精度泄漏检测，满足天然气泄露检测准确性的要求，且模型具有较好的泛化能力和鲁棒性。

Natural gas leak detection based on U2 Net and VGG16

At present,there is little research in the field of deep learning using natural gas leakage images as data.In this paper,a convolutional neural network(VGG16) trained on the methane leak images are used to realize automatic leakage detection.In addition,in view of the similarity between the leaked methane gas and the background image,the U2 Net image segmentation network is used instead of background modeling methods to extract the leaking gas region.By migrating the VGG16 network model structure and convolutional layer parameters,a BN layer is added between the convolutional layer and the excitation layer to improve the training speed,and the last pooling layer is replaced with a dynamic adaptive pooling method based on max pooling algorithm to improve detection accuracy.The improved VGG16 neural network is trained on the segmented infrared images and compared with other convolutional neural networks to evaluate the model using accuracy,precision,recall and F1 score,which performs best.Compared with the existing detection methods,the proposed detection method has higher accuracy.The detection method can achieve high precision leak detection,meet the requirements of the accuracy of natural gas leak detection,and the model has good generalization ability and robustness.