A Transfer-learning Based Defect Identification Method for Underwater Buildings in ROV
Underwater building defect detection is crucial to ensure the long-term safe and stable operation of the power plant.In order to solve the heavy detection task and elimimate the risk,improve detection efficiency,and reduce manual detection costs,a ROV-based underwater building defect identification method is proposed.Aiming at the problems of complex underwater imaging environment,high noise,and long detection process,an image recognition model based on transfer learning is designed.Firstly,the image data processing algorithm is employed to improve the quality of the underwater defect image,and the image is binarised to highlight the defect features.Then the salient features in the image are extracted by the convolution kernel,and the attention mechanism is introduced to calculate the allocation of feature importance,which improves the efficiency of the model feature extraction.Finally,migration learning is introduced into the training process to address the problem of insufficient actual defect data and improve model training efficiency.The results show that the proposed transfer learning image recognition model can accurately and efficiently identify underwater building features with an accuracy of 90%～95%on both standard and measured defect datasets,and the number of iterations is within 30 generations.
NETL
NSTL
万方数据
