Research On Automatic Detection Algorithm Of Fractures Around Wells Under Small Sample Conditions

With the rapid development of social economy and the continuous growth of energy demand,the exploration of fractured oil and gas reservoirs is of great significance.The fracture characteristics in imaging logging are more obvious and clearer than conventional logging data,which makes the imaging logging images become the main data for fracture detection.However,the workload of human-computer interaction for fracture detection is huge,simultaneously drilling and exploration engineering is vast,and it's very difficult and time-consuming to collect enough available imaging logging fracture images.So,this thesis is devoted to researching the automatic detection algorithm of fractures in imaging logging images with a small amount of fracture image data,that is,under the condition of small samples.In order to improve the efficiency of fracture detection,based on the ultrasonic logging images and according to the connectivity characteristics of fractures in the logging images,an algorithm combining threshold segmentation and ant colony algorithm was designed in the thesis--combinational optimal path search strategy.This algorithm performs threshold segmentation and uses the voting accumulation mechanism to establish a search space for fracture,then the fracture information in each sub-search space is recorded synchronously to realize the fast automatic detection.Results show that the proposed algorithm can avoid the search process from falling into local optimal solutions,the non-fracture information is filtered to a great extent,and the convergence speed is accelerated.Compared with the conventional ant colony algorithm,the time consumption is reduced by more than 98%,at the same time,it also shows a good recognition accuracy rate,which helps the logging interpreters to interpret the logging data well.Aiming at the shortcomings of the combinational optimal path search strategy such as single information,low recognition of disconnected fractures,as well as the lack of logging image data,and the shortage of labeled data sets.The thesis proposed an unsupervised domain transfer segmentation network based on convolutional neuralnetwork,it can realize automatic fracture detection from the smallest unit pixel level of the image.Its core is to use adversarial learning,including generators and discriminators,to promote the model to generate fracture segmentation similar to the source domain in the target domain.And an attention module is introduced to suppress the redundant noise in semantic segmentation results.The results show that the model has good results in terms of accuracy and visual quality when the logging image data is insufficient.Meanwhile,the trained network model has the advantages of fast speed,good robustness,and strong anti-interference ability in identifying and extracting fractures around the well.The algorithm in the thesis can use the computer to automatically identify the fractures in the logging images,reducing the workload of man-machine interactive fracture detection;The idea of synchronization and parallelism greatly improves the speed of fracture detection;The method of domain transfer enables the identification and extraction with high accuracy even in the absence of data.The works of the thesis provides a useful direction for the rapid fracture detection and fracture detection in the case of insufficient logging data.