Research On The Classification And Identification Of Microscopic Residual Oil Reservoir Storage Based On Machine Learning

Based on the knowledge of the remaining oil in the pore space of oil reservoirs,which is the basis for the deep development of oil fields and the guidance of residual oil exploration,this paper starts the research on the automatic classification and identification of the microscopic residual oil distribution pattern.However,the traditional microscopic residual oil distribution pattern classification recognition relies on professional experience judgment,and the judging standard does not have objectivity.And it takes a lot of time and effort to complete the labeling and classification recognition of residual oil in multiple images.To address this problem,this paper designs and implements a microscopic residual oil fugitive morphology classification and recognition system using image processing,image segmentation and classification recognition techniques.The details of the research are as follows:1.To address the problem of noise in the acquired microscopic residual oil distribution images,a microscopic residual oil distribution image denoising method based on improved bilateral filtering is proposed.The method increases the bilateral filter threshold,and improves the grayscale kernel function used in traditional bilateral filtering to minimize the influence of irrelevant pixel points on edge pixel points,which effectively achieves the purpose of protecting edges and removing noise.Besides,the saturation enhancement algorithm is used to enhance the microscopic residual oil distribution image.And the method converts the microscopic residual oil distribution image from RGB color space to HSL color space and sets the saturation increment to adjust the upper and lower limits of saturation so as to highlight the residual oil color information in the image more.2.To address the problems that the threshold value of the traditional Sobel operator is not adaptive and the calculation direction is relatively single,a method of residual oil edge detection based on the improved Sobel operator is proposed.Firstly,the method uses the Otsu algorithm to obtain the optimal threshold of the microscopic residual oil distribution image adaptively.And then it uses the Sobel operator to calculate the amplitude of each gradient of the image in the four directions after gradient enhancement.Finally,the non-maximum suppression algorithm is used to filter out the maximum value points and filter out the pseudo-edge pixel points in the four gradient directions to obtain the residual oil edge detection image.This method can detect the residual oil edge information in the microscopic residual oil distribution image more accurately and has strong noise immunity.3.To address the problem of low efficiency of artificial discrimination of microscopic residual oil reservoir storage forms,a classification and identification method of microscopic residual oil reservoir storage forms based on weighted random forest is proposed.The method first calculates separately the voting error rate of the decision trees in the weighted random forest for different residual oil pattern categories and then the weight of each decision tree to different morphological categories is updated according to the voting error rate.Secondly,the voting weights are multiplied by the voting results of the decision trees,and the weighted votes of each category are summed.Finally,the classification result with the largest number of votes is used as the result of classification recognition of the residual oil reservoir storage forms.This method improves the classification performance while maintaining stable generalization performance.4.The system is designed and implemented to meet the needs of classification and recognition of microscopic residual oil reservoir storage forms.The system functions mainly include bilateral filtering and denoising,saturation enhancement,image edge detection and classification,and recognition of residual oil reservoir storage forms.