无线电工程

水下建筑物缺陷检测是保障电厂长期稳定运行的关键任务之一。针对水下建筑物能见度低、人工检测成本高、检测任务危险、缺陷检测精度低等问题，提出一种基于改进Mask R-CNN的水电站水下建筑物缺陷检测算法。采用图像处理技术提高水下缺陷图像质量；通过K均值聚类算法确定先验边界框，提高模型效率；在网络中添加注意力机制，聚焦重要信息，增强网络对水下缺陷的关注度，提高模型性能和检测的准确度；修改特征融合网络为SS-FPN(FPN Scale Sequence),减少特征融合时信息丢失，加强语义融合。对比试验结果表示，与改进前基于ResNet50的Mask R-CNN算法相比，改进后的算法提高了水电站水下建筑物缺陷检测的精度，后续处理得到的缺陷轮廓更精确。

Defect detection of underwater buildings is one of the key tasks to ensure the long-term stable operation of power plants. To solve the problems of low visibility of underwater buildings, high cost of manual detection, dangerous detection tasks and low accuracy of defect detection, an underwater building defect detection algorithm based on improved Mask R-CNN is proposed. Firstly, image processing technology is used to improve the quality of underwater defect images; then K-mean clustering algorithm is used to determine the a priori bounding box to improve the efficiency of the model; the attention mechanism is added to the network to focus on the important information, enhance the network's attention to underwater defects, and improve the performance of the model and the accuracy of detection; and the feature fusion network is modified to SS-FPN(FPN Scale Sequence) to reduce the loss of information in the fusion of features and to enhance semantic fusion. Comparison test results indicate that compared with the ResNet50-based Mask R-CNN algorithm before improvement, the improved algorithm improves the detection accuracy of defects in underwater buildings of hydropower stations, and the defect contours obtained from subsequent processing are more accurate.