基于CBAM-Res_UNet电厂高压蒸汽泄漏检测研究

发电厂高压蒸汽泄漏检测关乎电厂设备长期稳定运行。 为了提高电厂高压蒸汽泄漏检测的准确性,解决泄漏区域的错 分割和漏分割问题,提出基于 CBAM-Res_UNet 图像分割网络的电厂高压蒸汽泄漏检测算法,在 UNet 结构中加入 ResNet 的残 差块 residual_block 来获取泄漏图像更多的语义信息,并且融入 CBAM,加强高压蒸汽泄漏图像区域特征的学习,网络再根据不 同损失函数和评价标准对图像分割结果的影响,选择损失函数 Focal Loss+Dice Loss 和性能指标 F1_score。 通过在电厂高压蒸 汽泄漏图像数据集上进行实验,CBAM-Res_UNet 网络得到的 F1_score 值为 0. 985,实验结果表明,该网络可以更加完整的分割 出蒸汽泄漏区域,对高压蒸汽泄漏图像多样性有较强的泛化能力。

Research on leakage detection of high pressure steam in power plant based on CBAM-Res_Unet

The detection of high pressure steam leakage in power plant is related to the long-term stable operation of power plant equipment. In order to improve the accuracy of high-pressure steam leakage detection in power plants and solve the problem of wrong segmentation and leakage segmentation of leakage areas, this paper proposes a high-pressure steam leakage detection algorithm based on CBAM-Res_UNet image segmentation network. The residual_block of ResNet is added to the UNet structure to obtain more semantic information of leakage images, and CBAM is integrated to strengthen the learning of regional characteristics of high-pressure steam leakage images. According to the influence of different loss functions and evaluation criteria on image segmentation results, the loss function Focal Loss+Dice Loss and performance index F1_score are selected. Through the experiment on the image data set of highpressure steam leakage in power plant, the F1_score obtained by CBAM-Res_UNet network is 0. 985. The experimental results show that the network can more completely segment the steam leakage area, and has a strong generalization ability for the variety of high pressure steam leakage images.