| The increasing demand for oil consumption has put forward new requirements for logging technology,especially the application of artificial intelligence technology in logging interpretation tasks,which has brought about a new round of technological revolution.In practice,the intelligent algorithm needs to first solve the problem of regional division and identification of geological features of well logging images.Therefore,this study is greatly inspired by the semantic segmentation technology that has been widely concerned in other fields.In this study,the geological feature extraction and intelligent recognition algorithm of electrical imaging logging image are designed based on the problems of regional discontinuity and target mis-segmentation of U-Net algorithm based on encoder-decoder framework in the segmentation of electrical imaging logging image.The main contents are as follows.(1)To address the problems of incoherent regional in the segmentation of electrical imaging logging images due to low feature utilization and lack of connection between features at different levels in the U-Net network,the SV-UNet network is considered to be proposed by incorporating the Shuffle Net V2 module in this network to enhance the learning and expression capabilities of the model and to achieve the segmentation of geological feature regions in images;(2)In order to further improve the segmentation performance,for the SV-UNet network,due to the semantic difference between the same group of features in the skip connection layer and the loss of spatial information in the downsampling process,the target mis-segmentation problem exists in the segmentation of electrical imaging logging images.The concurrent spatial channel Squeeze and Excitation(sc SE)module is introduced in the skip connection layer of the network,followed by the Pyramid Pooling module in the final encoder layer,and finally the Weighted SP-UNet network is proposed to make full use of the global context information,reduce the misjudgment rate of the model,and improve the accuracy of the segmentation of geological feature regions in the image.Experiments were carried out on the electrical imaging logging image data set containing six types of different geological features.The average Pixel Accuracy(PA)of the Weighted SP-UNet algorithm reached 97.09%,and the comprehensive mean Intersection-over-Union(m Io U)reached 83.53%,which are 1.29% and 4.39% better than U-Net.The experimental results show that the proposed algorithm can effectively improve the segmentation effect,has good regional coherence and target discrimination,and can assist the logging staff in more accurate imaging logging interpretation and reduce the difficulty of logging interpretation in practical application. |
