Research On Image Classification And Video Mosaic Of The Axial View Panoramic Borehole Camera System

The structure characteristics and distribution of structural planes,such as fracture,abscission layer,broken rock zone and interface,are the key factors that determine the geological features and mechanical properties of geotechnical engineering in deep strata.It plays an important role in studying the stability of geological structure,rock engineering design and construction safety.The borehole videos and images can be obtained by the axial view panoramic borehole camera system(APBCS),which is an important means of geological survey in the observation and analysis of the borehole inner wall structure.Investigating the effective methods to use the video and image data to study the geological borehole and further improve the performance of the APBCS becomes an important research topic.Currently,there are two problems in the APBCS:Firstly,the interpretation of the borehole image acquired by the APBCS is mainly based on the manual operation.The borehole image datasets are huge while human involvement is time-consuming and laborious.Similarly,different engineers arrive at different conclusions.In addition,the APBCS technology is a way that engineers survey the condition of the borehole inner wall based on the video playback and video screenshot.However,the borehole wall unfolded image cannot be acquired by the APBCS technology.It can only provide qualitative description for the structural plane,but not implement quantitative analysis for the direction,width and inclination angle of the fracture.To address the problems above,based on the borehole image and video acquired by the APBCS,the recognition and classification method of the forward looking borehole wall image is studied,and the mosaic technology of forward looking borehole wall video is also investigated in this dissertation,acquiring the cylindrical panoramic borehole wall unfolded image by the borehole center location algorithm,the image unwrapping algorithm and the rectangular unfolded image sequence mosaic technique.The main contents of this dissertation are as follows:A new two stage tree image classification method is proposed in this dissertation,which has realized the identification and classification of the three main types of borehole wall images by optimizing a two stage feature extraction structure.There are two stages in the method:primary classification and secondary classification.The primary classification process is to make a binary decision on the three types of images and to decide whether they belong to the border image or non-border image.In the secondary classification process,a fracture segmentation algorithm based on the Gabor filter and a central region removal algorithm based on circularity are proposed to construct more dissimilarity eigenvector as an input to the secondary classifier.The experimental results show that the different categories of the borehole wall image can be effectively identified by the two stage classification method with the correction rates of 93.3%in the test set.Research results could provide theoretical reference and new ideas for the classification of the forward looking borehole wall image.A cylindrical panoramic borehole wall unfolded image generation method based on the video mosaic is presented in this dissertation to enhance the performance of the APBCS from the qualitative observation to the quantitative analysis in aspect of borehole detection.In the method,the axial motion parameter estimation is converted to the translational motion parameter estimation with the image unwrapping algorithm.The rectangular borehole wall unfolded image sequences are then stitched together as a segmentation or integration of cylindrical borehole wall unfolded image.In order to eliminate the geometric distortion and confirm the annular effective unwrapping region during the image unwrapping process,a borehole center automatic location algorithm is proposed.The borehole center of different video frames can be obtained accurately according to the automatic segmentation and calibration of central region.Subsequently,an unwrapping algorithm of forward looking borehole wall image is proposed to unfold the annular effective region in the video sequence into the rectangular borehole wall unfolded image sequence.A virtual coordinate map based on the coordinate mapping algorithm is designed and combined with the bilinear interpolation to meet the unification of image resolution in inner and outer ring regions and to increase processing efficiency of the video sequences.According to the problem that the registration accuracy of the template matching algorithm is poor during the mosaic process of the borehole video,a projection transformation sum of absolute difference algorithm(PTS AD)is proposed,which uses the gray projection transformation(GPT)to decision-making of the candidate images selected by the sum of absolute difference(SAD).Afterwards,the optimal candidate image matched with the template image can be determined by the projection correlation coefficient(PCC)and used for fast image sequence matching.Thus,the horizontal and vertical offsets of the unfolded image sequence can be calculated by using the relative offset distance of center coordinates between template image and optimal candidate image.The rectangular unfolded image sequences are then stitched together as one single cylindrical panoramic borehole wall unfolded image using these two parameters.Experimental results demonstrate that the proposed PTSAD algorithm has a good performance in accuracy and efficiency compared with the feature-based matching methods and with the pixel intensity-based matching methods,which can meet the real-time and effectiveness of the video mosaic process.Aiming at the invalid video frame caused by the vibration and rotation of the probe,a video frame filtering module is designed to remove the invalid video frame before the mosaic of rectangular unfolded image sequence by using three image parameters such as the central offset dc,the horizontal offset dh,and the vertical offset dv.Experiments demonstrate that the proposed cylindrical borehole wall unfolded image generation method can produce panoramic unfolded images in different kinds of forward looking borehole videos with satisfying visual effect,and it can provide the research basis for the quantitative analysis of fracture,bedding,joints and other structural planes in borehole.