Crack detection on dam surfaces is an important task for safe inspection of hydropower stations. More and more object detection methods based on deep learning are being applied to crack detection. However, most of the methods can only achieve the classification and rough location of cracks. Pixel-level crack detection can provide more intuitive and accurate detection results for dam health assessment. To realize pixel-level crack detection, a method of crack detection on dam surface (CDDS) using deep convolution network is proposed. First, we use an unmanned aerial vehicle (UAV) to collect dam surface images along a predetermined trajectory. Second, raw images are cropped. Then crack regions are manually labelled on cropped images to create the crack dataset, and the architecture of CDDS network is designed. Finally, the CDDS network is trained, validated and tested using the crack dataset. To validate the performance of the CDDS network, the predicted results are compared with ResNet152-based, SegNet, UNet and fully convolutional network (FCN). In terms of crack segmentation, the recall, precision, F-measure and IoU are 80.45%, 80.31%, 79.16%, and 66.76%. The results on test dataset show that the CDDS network has better performance for crack detection of dam surfaces.

Underwater structure inspections are essential for infrastructure maintenance, such as hydraulic facilities, bridges, and ports. Due to the influence of turbidity, dark light, and distortion, the traditional methods cannot satisfy the requirements of on-site inspection applications. This paper proposed a methodology of the point cloud data capture in the turbid underwater environment. The method consisted of an acquisition device, a distortion correction algorithm, and a parameter optimization approach. The acquisition device was designed by composing a silt-removing module, a structured light camera module, and a clear water replacement module, which can integrate with an underwater inspection robot. The underwater multi-medium plane refraction distortion model was established through analysis, and a refraction correction algorithm was provided to correct the distortion. To obtain the maximum field of view of the point cloud, the nonlinear optimization approach was used to select the medium material and thickness. After the real experiments using the Intel RealSense sr300 depth camera, maximum measuring distance could range up to 253 mm in water, the accuracy of the point cloud of the underwater target objects was ±3.77 mm, and the maximum error was 8.76%. Compared with other methods, this method was more suitable for 3D point cloud capture in the turbidity environment.

When acquiring object point cloud data by three-dimensional scanning technology, noise is generated due to instrument accuracy and external factors. Existing algorithms rarely consider the characteristics of different noises and different regional noises when solving the point cloud denoising problem, this results in a limited effect on denoising. This paper presents an algorithm for denoising based on the characteristics of different types of noise and different regions in the point cloud. The algorithm includes large-scale noise removal and small-scale noise smoothing. Remove large-scale noise points by the relationship between the local point cloud and the global point cloud. For small-scale noise, the feature regions and non-feature regions are extracted according to the normal cosine information entropy. According to the characteristics of the small-scale noise in two regions, the noise distance distribution and the optimized bilateral filtering are used to deal with the small-scale noise in two regions respectively. Comparison experiments show that our algorithm can effectively remove the noise points mixed in the normal point cloud. The accuracy of large-scale noise removal reaches 99.1%. The proposed algorithm can protect the feature areas from being over-smoothed while smoothing the small-scale noise in non-featured areas.

Exposed aggregate is a typical feature of the abrasion erosion in stilling basin slabs concrete surface. Although a variety of underwater robots are designed for inspection, the exposed aggregate detection for identifying abrasion is often done by manual work. The scarcity of image samples, large differences in aggregate size, color and shape are the main difficulties in automatic detection. To address this problem, an improved Attention U-Net deep fully convolutional network-based detection method was proposed. To realize this method, underwater images in site were captured via a self-developed operated underwater robot. Through randomly separating and the cropping of the 128 underwater images, the 512×512 pixels images dataset was built according to the ratio of 8:1:1, including 408 training images, 52 validation images and 52 test images. After the data augmentation, loss function and the optimizer were carefully designed and selected, the proposed Attention U-Net architecture was evaluated on this dataset. For comparative research, the full convolution network (FCN) and U-Net network were trained with the same training and validation dataset. The performance comparison on the test dataset showed that the Attention U-Net architecture has better detection accuracy.