TPDNet: A point cloud data denoising method for offshore drilling platforms and its application.

[Display omitted] • A deep learning model of TPDNet based on the FA module and the self-attention mechanism is proposed, which can effectively identifies and separates noisy and target points from offshore drilling platform point cloud data. • The datasets were constructed using a combined filtering algorithm combining KNN_PCF and Hy_WHF to separate the outlier points, and the noise points of the cluttered objects point cloud (such as cluttered lines) are extracted with manual labeling. At the same time, in order to balance the ratio of the noise, Gaussian noise is introduced to increase the number of outlier points. • An application of a surface reconstruction method for the point cloud data of an offshore drilling platform based on the deep neural network using contextual prior information is implemented. The complex working environment of offshore drilling platforms makes the acquisition of point cloud data susceptible to noise pollution. To address this issue, this paper proposes a denoising network for point cloud data of offshore drilling platforms, called TPDNet. TPDNet utilizes the feature abstraction module to aggregate local features in point clouds and employs a self-attention mechanism for feature extraction, thereby enabling the effective identification of noisy point clouds. This paper also presents an offshore drilling platform point cloud dataset for training and testing deep learning models. It demonstrates the reconstruction of 3D surfaces of equipment on offshore drilling platforms using the target point cloud data obtained by TPDNet. The result validates the practicality of TPDNet. Consequently, this paper provides technical support for point cloud data processing, which has promising practical applications in the field of ocean engineering. [ABSTRACT FROM AUTHOR]