Research On Image Enhancement Algorithm And Design Of Host Computer Software Of Borehole Ultrasonic Imager

In the field of petroleum logging,borehole imaging logging is an important means to evaluate oil and gas Wells because it can generate intuitive logging image with fracture and hole information to reflect the condition of oil and gas Wells.The main logging methods used in borehole imaging logging are ultrasonic imaging method and micro-resistivity scanning method.Among them,ultrasonic imaging method has been widely used because of its strong penetration ability,high borehole coverage rate and simple instrument structure.In this logging method,ultrasonic signals are sent to the borehole wall,and the arrival time and the amplitude of the echo signal is detected.The detected data is uploaded to the host computer software to draw the borehole wall image.Because of the downhole environment and the limitation of hardware,the quality of logging images obtained by the ultrasonic imager usually is poor,which is not conducive to the interpretation of logging personnel.The research of the thesis mainly consists of the following aspects:（1）To improve the quality of logging image,aiming at the problems of low contrast and unclear details in logging image,this thesis designs an enhancement algorithm of logging image based on the deep learning—super-resolution reconstruction algorithm based on high frequency feature enhancement.Multi-scale cavity convolution block is designed as the basic module of the network to extract features of different scales,and the high frequency features are separately trained.Finally,the experimental results show that compared with other classical algorithms which tested in logging image dataset,the proposed algorithm has good objective index and subjective visual effects.（2）In order to implement the algorithm in practical engineering,the large computation and large storage capacity of the super-resolution reconstruction model based on high-frequency feature enhancement need to be solved.The thesis proposes a model compression algorithm based on knowledge distillation,which distills two different student models based on tutor-student architecture,and then lets the two student models learn from each other to enhance their generalization.Experimental results show that this algorithm reduces the model size to nearly one ninth of the original model,and the objective index is improved.（3）For the sake of interacting with the downhole logging circuit,the thesis uses UML language for analysis,and uses Labwindows/CVI platform to design the user interface and function modules of the host computer software of circumferential borehole ultrasonic imaging logging tool,which has the functions of communication,parameter setting,online download,data processing,image display,file storage and so on.Finally,the software is tested jointly with the downhole logging circuit.The experimental results show that the software has complete functions and is convenient to use.It can directly display the high-quality logging images and is convenient for logging personnel to analyze.