Research On Lithology Identification Method Of Cuttings Image Under PDC Bit Condition

Cuttings logging is an integral part of the current petroleum exploration process.In terms of geological evaluation,the geological structure and oil and gas properties are directly reflected by cuttings logging.In terms of the safety of oil and gas exploitation,the geological parameters can be obtained by analyzing the relevant information in cuttings logging,and the hidden dangers in engineering can be found in advance.In the current drilling technology,PDC bit has been widely developed and applied in the exploration field due to its advantages of high length,high rotational speed and low drilling cost.However,the cuttings obtained by drilling with PDC bit are very small and small,which will reduce the accuracy of artificial identification of cuttings and affect the identification of lithology of cuttings.Therefore,it is of great market value to design a cuttings image segmentation and recognition method that can effectively,quickly and accurately replace human to complete cuttings logging.The segmentation and recognition method of cuttings image are mainly studied.In image segmentation,an instance segmentation method based on deep learning is proposed.In the aspect of image recognition,a lithologic identification method based on oil-bearing lithologic feature database is proposed.The main work of this paper is as follows:Firstly,the image needs to be preprocessed before segmentation and recognition.Aiming at the problem that there are a lot of extreme noise in the cuttings image,morphological processing is carried out on the cuttings image to eliminate extreme noise.Based on the morphological theory,the morphological geodesic dilation and geodesic corrosion are used to reconstruct the cuttings images.Secondly,in order to solve the limitation of traditional image segmentation algorithm in the segmentation accuracy,deep learning correlation algorithm is studied in the cuttings image segmentation.The instance segmentation task is more complex than the semantic segmentation task.The general deep learning method has a good effect on semantic segmentation,but in instance segmentation task,it has a poor effect on class segmentation.U-Net++ instance segmentation method based on multi-task learning is proposed.In this model,a complex task is divided into two sub-tasks,and the two tasks are trained and learned by the same network.Parameters of two tasks are shared in shallow layer,which simplifies complex tasks while improving learning efficiency and reduces the risk of single task falling into local optimum.Furthermore,in order to better fuse semantic information and edge information,a multi-feature fusion method based on superpixel optimization is proposed.U-Net++ instance segmentation method based on multi-task learning for semantic segmentation results and edge segmentation results,after super pixel algorithm are used to get the pixel sub-block,with characteristics of optimal fusion method,the image segmentation result instance is improved,the segmentation accuracy is improved,the problem of the direct fusion appear too much noise is solved,Experiments show that cuttings can be effectively separate.Thirdly,in the image recognition problem,using the advantage of better expression of color in HSV space,the location of oil-bearing cuttings is accurately located by fluorescence threshold segmentation algorithm,and the texture and color features of oil-bearing cuttings are extracted to form the oil-bearing feature library.In order to solve the problem that the cuttings formation and the cuttings image acquisition process are affected by external factors,the similarity degree of the cuttings characteristics and the features of oil-bearing reservoir is improved by adaptive Bhattacharyya distance calculation,and the lithology identification of oil-bearing cuttings is completed.Experimental results show that this method has the advantages of high efficiency and high precision,and has certain robustness.Finally,the cuttings images collected in a certain area are segmented and recognized,and the area proportion of oil-bearing cuttings in the cuttings map is calculated.Experimental results show that the error of this method is less than 5% in each well depth,the time is much lower than manual identification(300s),and the identification effect is better than 110 s.