Research On Ultrasonic Image Stitching Of Material Defects Based On Data Feature Fusion

Ultrasound detection and imaging technology is widely used in material defect detection,providing information on the size and location of defects,which serves as a basis for defect detection and overall material performance evaluation.However,during a single ultrasound transducer scan,the scanned area is limited,and the transducer must be continuously moved to achieve overall detection of large components or materials.After obtaining the echo information of the detection area at each detection position,the transducer completes defect detection and imaging based on the processing of the echo information.When a component is large,it may be impossible to obtain a panoramic distribution of defects for the entire component in a single scan.Imaging multiple distributed small or large defects can sometimes cause artifacts that can impact the overall evaluation of the material or component’s performance.This paper proposes a data stitching method based on the characteristics of multiple ultrasound detection echo signals.The method directly processes the obtained echo signals and stitches together the continuous defect areas to obtain the panoramic distribution of component defects.This method eliminates the need for image stitching based on image features after imaging,thereby reducing the image stitching process and improving the efficiency of ultrasound image stitching.The thesis first analyzes the current status and characteristics of existing ultrasound imaging methods and analyzes the current status and technical characteristics of non-ultrasound image stitching.Based on this,the ultrasound image stitching method based on image features is described,and the applicable occasions,technical features and limitations of the current image stitching methods are summarised and summarised.The noise in the echo signal is filtered out,and the artifacts caused by signal trailing and other phenomena that affect the stitching quality.is removed by proposing So S function-based artifact removal method for ultrasound detections data.The echo signal data after removing the artifacts is used as the data to be stitched,and a method for registering ultrasound echo data based on normal vectors is proposed.Firstly,feature points are selected based on the local normal vector changes of ultrasound data,and the feature points are described using their histograms.The feature points are roughly matched using the Euclidean distance comparison criterion,and then fine matching is performed using the adaptive threshold random sample consensus algorithm.Finally,the registration parameters are calculated using the quaternion method to complete the data registration.Then,an average fusion method based on KD-tree is proposed to fuse the registered data and obtain an ultrasound panoramic image.The stitched image is evaluated using image stitching evaluation indicators through practical testing.The main research work and conclusions of the thesis are as follows:1)The current status of research on image-based and data-based stitching methods in different fields is analysed and summarized.The purpose and significance of studying ultrasonic panoramas of material defects is described based on the characteristics of defect ultrasound detection and imaging in industry.Additionally,the factors that affect the quality of image stitching are analyzed,and the current status of research on artifact rejection methods in ultrasound images is examined.2)The commonly used ultrasonic defect detection and imaging principles are introduced.The general idea and scheme design of the research on ultrasonic image stitching of material defects based on data feature fusion is given.The design and construction of an ultrasonic defect detection system are described,including the functions and principles of its components.To test the technical feasibility of the system,standard test blocks for defect detection are designed and prepared,and actual testing is conducted.The resulting detection echo data is stitched together.This analysis provides a foundation for subsequent data processing and image stitching based on the data.3)The mechanism of artifact generation affecting ultrasound image quality is analysed,and the effect of artifacts on ultrasound image stitching is also analysed,and the artifact removal method based on the So S kernel function for ultrasound detection data is proposed.This method effectively reduces the degree of significance of artifacts by pre-processing the detection echo data.To evaluate the proposed method,ultrasonic echo signals of standard defect specimens of slot and through-hole types were collected and convolutionally modulated.The detection echo data were obtained using a data sparse sampling method based on finite rate of innovation.Subsequently,the echo time delay and amplitude parameters were estimated from the sampled data by an annihilating filter algorithm,and the echo signals were reconstructed as data for image stitching.The effect of artifact removal was verified by direct imaging of the reconstructed signal.4)A normal vector-based alignment method for ultrasound echo data is proposed to address the problems that the amount of information in ultrasound echo data is richer than that in ultrasound images,and that the stitching process is more tedious when using image pixels for feature extraction and stitching.The method involves selecting feature points based on the variation of local normal vectors of ultrasound data and describing them using a histogram of feature points.Coarse matching of feature points is performed using the Euclidean distance comparison criterion,followed by fine matching using the random sampling consensus algorithm with adaptive thresholds.The alignment parameters are then calculated using the four-element method,and different data are converted to the same coordinate system to achieve alignment of ultrasound echo data.5)The processing of aligned repetition points by point clouds was analyzed,and the KD-tree based averaging fusion method was proposed for the aligned ultrasound echo data.The KD-tree nearest neighbour search can be used to identify the overlapping data between two ultrasound echoes after alignment.The overlapping data is then added and averaged directly,while the non-overlapping parts remain unchanged,to complete the fusion of the ultrasound echo data.This process results in the fusion of ultrasound data,which is then imaged to obtain an ultrasound panorama.6)According to the criteria for evaluating ultrasound image stitching,the databased image stitching algorithm proposed in the paper is validated and evaluated.The results show that after pre-processing,feature extraction,data alignment,fusion and imaging of the ultrasonic transducer detection data,a panoramic view of the defect distribution of the specimen can be obtained from the detection data obtained from multiple sweeps of the transducer,with a stitching success rate of more than 99%.The stitched ultrasound image can effectively detect defects and visually observe their size and distribution,providing a reference and basis for the overall defect assessment of the material or component.