基于块分解和循环神经网络的海上升压站智能巡检目标时序配准算法OACSTPCD

The timing registration of intelligent inspection targets for conventional offshore booster stations mainly relies on triangular grid representation methods,but it lacks analysis of the energy distribution of image blocks,resulting in a large proportion of outliers after target registration and poor registration results.To this end,a time-series registration algorithm for intelligent inspection targets of offshore booster stations based on block decomposition and recurrent neural networks is proposed.By analyzing the inspection items of offshore booster stations,the block decomposition algorithm is used to decompose the inspection images,detect the feature points of inspection targets,and extract the high-frequency components in the decomposed image blocks to obtain the energy distribution of images.Combined with the main curvature of target feature points,the scalar scaling coefficient of feature points is calculated,and compared with a given threshold to identify unstable extreme points and remove them.Based on this,a recurrent neural network is introduced to obtain the standard deviation of pixels between the centroid of inspection target point and the feature points,in order to identify the center point of the array for registration.Taking a practical offshore booster station project as the research background,the performance of the proposed method is verified.The results of the example show that,after using the proposed algorithm for temporal registration of inspection targets,the proportion of outliers is smaller and the registration effect is better.