| To ensure that the well remains in good initial shape over long periods of time,the casing is used to support the wellbore after drilling.The casing pipe will be damaged in different degree after long use,so it is necessary to check it.At present,the casing damage image recognition obtained by video logging technology is generally manual recognition,which is strictly professional requirements,highly degree of subjective participation,large workload.The work of this paper is to find a more efficient algorithm of casing damage identification based on machine learning,which can reduce workload and work difficulty(1)The image was denoised using the Gaussian filtering method,and the image segmentation method was optimized based on the adaptive threshold algorithm to optimize the traditional Canny operator’s image segmentation method.A sample bank of casing damage images was constructed.(2)After extracting the grayscale features,texture features and shape feature parameters of the casing damage image,the BP neural network and genetic algorithm optimize the BP neural network were used to identify the casing damage image.(3)ResNet and VGG16 networks were trained respectively based on the convolutional neural network structure.A new network model was constructed by using the Stacking model integration method in combination with ResNet and VGG16 networks,and training of casing damage image recognition was performed.Experiment found using VGG16 network and ResNet network get accuracy were 79.3% and 83% respectively,and the method based on Stacking model integration achieved 86.5% accuracy,improve the accuracy,and use genetic algorithm to optimize the BP neural network model after the network convergence speed than BP neural network model has obvious improvement,and the recognition accuracy rate from 81.6% reached 85%.Through the training of five models,it is shown that based on machine learning,a good recognition effect can be obtained on the casing damage image. |
PDF Full Text Request