Image Recognition Of Oil-bearing Cuttings Based On Deep Learning

In the process of petroleum exploration,cuttings logging provides a basis for studying geological structure,analyzing hydrocarbon characteristics and establishing geological models.Polycrystalline Diamond Compact(PDC)bit is widely used in cuttings logging,but the cuttings captured are fine and close fitting,the surface color of cuttings is dark,the cuttings boundary is blurred,which affects the efficiency of follow-up work directly.Cuttings logging is usually done manually to collect,extract and analyze.This will lead to greater subjective factors,which will result in the deviation of subsequent data and analysis conclusions,thus it will affect the judgment and identification of experts.Therefore,it is of certain significance to study the segmentation and recognition of the cuttings image obtained by PDC bit.Due to the low accuracy of traditional image segmentation methods,it is difficult to recognize and classify cuttings in the later stage.In this paper,a deep learning based image recognition method for oil-bearing cuttings is proposed.Firstly,the cuttings image is denoised and the image contrast is enhanced.In order to solve the problem that the scene of cuttings image is too complicated,a combination of Gaussian filtering and bilateral filtering technology is used to denoise the cuttings image.In order to solve the problem of low brightness and low contrast of cuttings image,histogram equalization method is used to obtain suitable cuttings data for testing.Secondly,an improved Multi Res-UNet++ image semantic segmentation algorithm is proposed to realize chip image segmentation.Multi Res-UNet++ not only retains the characteristics of Inception module,but also extends the residual connection.It can segment the cuttings image quickly and accurately.In order to extract more effective features of cuttings,enrich convolutional layers and dense blocks,the improved Multi Res module is integrated into U-net++ network,which can fully extract and utilize features of cuttings image,it also avoid gradient disappearance and short training time.The characteristic data of the input network is processed in batches,and BN layer is added before each input layer of the basic U-net++ network,which can effectively improve the generalization ability of the network,so that the data distribution of each layer is stable when the network is trained.Adding the convolutional attention mechanism combining Channel and Spatial into the network can effectively reduce the amount of network computation.According to the different importance of pixels,with the help of the convolutional block attention module,we can learn the adaptive allocation of feature weights,so as to improve the computing speed.The improved algorithm can not only improve the effect of network segmentation but also ensure the speed of network segmentation.The performance evaluation index reaches about0.8,and the single training time only takes 70 seconds.And then,An improved Res-Net34 combined with transfer learning algorithm was used to recognize the image of oil-bearing cuttings.The cuttings individual obtained by the segmentation algorithm is compared with the cuttings with the threshold value segmentation under the condition of fluorescence,and the cuttings individual is extracted,the high-quality image feature database dataset of the cuttings is established.A new method for oil-bearing cuttings recognition based on deep learning and transfer learning is designed.The improved Res-Net34 is used as the pre-training model for cuttings image recognition transfer learning.The network structure and pooling layer of Res-Net34 are improved.Not only the number of layers and miscellaneous parts of the network are simplified,the training convergence speed is improved,the accuracy of the network identification of oil-bearing cuttings is improved,and the feature extraction of the pre-training network is more accurate.At the same time,in order to prove the recognition performance of the improved algorithm,several simple neural network algorithms and the improved Res-Net34 algorithm were compared to carry out the cuttings image feature extraction experiment,and then the cuttings image open data set was selected for migration learning.The cuttings image feature extracted by the algorithm was retained as the initial weight,and fine-tuning parameters were used in the configuration method to realize the identification of oil-bearing cuttings.Finally,according to the whole segmentation and recognition system,a network is built and applied to the oil-bearing cuttings image under the condition of PDC bit.Firstly,the cuttings image under the condition of PDC bit is pre-processed,such as de-noising and contrast enhancement,and then the improved Multires-UN ++ algorithm is used for image segmentation.Then the improved Res-Net34 combined with transfer learning algorithm is used to realize oil content identification.The ablation comparison experiment of oil-bearing cuttings image identification with different sizes and original images is added.The test experiment verifies that the algorithm proposed in this paper can realize the oil content identification of basic cuttings image,and realize the automation and intellectualization of cuttings logging.