基于改进SSD算法的电厂设备油液渗漏检测

An Improved SSD Algorithm Oil Leakage Detection of Power Plant Equipment

摘要:

为能实现对电厂充油设备和管道的油液渗漏现象快速、准确的检测与识别，通过引入高分辨率网络实现高分辨率特征提取，改进特征融合模块以融合高分辨率特征信息强化特征表达，提出了一种基于改进SSD的油液渗漏图像检测算法。此外，针对油液渗漏现象构建一个电厂设备的油液渗漏数据集并提出了一种随机种子遮挡的数据图像增广策略。经实验测试表明，算法在检测效果上提升明显，相比于基于SSD算法的的漏油检测模型的准确率和召回率分别提高了3.1%和3.7%，满足了工程实际需求，具有较高的实用性。

Abstract:

To achieve rapid and accurate identification and detection of oil leakage in oil filled equipment and pipelines of power plant, an improved SSD oil leakage image detection algorithm is introduced in this paper. Firstly, high-resolution network is applied to achieve the feature extraction with high-resolution. Secondly, feature expression ability is increased through high-resolution feature information integration with modified feature fusion model. Moreover, a data image augmentation strategy is presented based on random seeds shading. Finally, a oil leakage dataset is constructed in a electrical factory equipment to solve the problem of oil leakage. Compared with SSD, verification experiment show that the accuracy and recall increased by 3.1% and 3.7% respectively. It improves the detection ability significantly, which meets the needs of engineering.