【摘要】 裂纹是水下结构物中最常见的缺陷之一，得不到及时检修可能会危害整体结构安全，造成重大事故。传统的裂纹检测方法费时费力且效率低下，文中提出了一种由水下机器人搭载目标检测算法的水下裂纹检测方法。算法基于YOLOX模型，在检测网络中引入自注意力机制和添加空洞空间卷积池化金字塔（ASPP）结构，同时改进激活函数，并制作了水下裂纹数据集。将算法在数据集上进行试验，结果表明，改进后的算法训练损失更低，模型收敛更快，AP值相比于原始YOLOX模型提升了2.07%，相比于YOLOv5提升了4.35%；使用不同大小的数据集进行试验，发现改进后的算法随着数据集的增大检测性能提升更快，更适用于大规模数据集；最后将算法应用于水下裂纹的检测取得了较为良好的识别结果。更多还原

【Abstract】 Cracks are one of the most common defects in underwater structures. Failure to repair them in time may endanger the safety of the overall structure and cause major accidents. Traditional crack detection method consumes a lot of manpower and material resources with low efficiency. Therefore, an underwater crack detection method based on the target detection algorithm of underwater robot deployment is proposed. Based on YOLOX model, self-attention mechanism and ASPP structure were introduced into the detection network, activation function was improved, and an underwater crack data set was made. The algorithm is tested on the underwater crack data set. The results show that the improved algorithm has lower training loss and faster model convergence. The AP value is increased by 2.07% compared with the original YOLOX model, which is 4.35%higher than YOLOv5. Experiments with data sets of different sizes show that the detection performance of the improved algorithm improves faster with the increase of data sets, and it is more suitable for large-scale data sets. Finally, the algorithm is applied to the detection of underwater cracks and obtains relatively good recognition results.更多还原