Research Of Ultrasonic Detection’s Signal Processing With Pipeline Corrosion Based On Support Vector Machines

Long distance pipeline corrosion detection is the key of pipeline safetydetection.Ultrasonic inner inspection is the main testing methods,ultrasonic inner inspection echo signal processing is the key technology ofultrasonic inner inspection of pipeline, it is performance analysis and defectrecognition based.By analyzing the principle of ultrasonic inner inspection, according tothe relations between the wall thickness of the pipe and the frequency ofthe corresponding, this article proposed methods in view of ultrasonicdetection signal processing feature extraction-A scan feature extractionmethod, frequency domain method, frequency domain feature extractionfeature extraction method, digital filtering aided method, analyzed andverified with the experiment one by one, improved and combinated on thebasis of experiment, finally put forward the method-improved frequencydomain feature extraction method, which are suitable for the subject;According to the corrosion types and distribution of experimental test thisarticle select binary tree multi-class support vector machine for ultrasonicecho signal of automatic classification. Algorithm use radial basis kernelfunctions, and in combination with improved frequency domain featureextraction method, The experimental results show that, the improvedfrequency domain feature extraction method for multi class classificationsupport vector machine are better performance in the classification accuracy and speed, achieved better classification results; Preliminarystudying of the crack flaw echo signals, and try to use support vectormachine to crack the qualitative analysis.Improved frequency domain feature extraction method make multi classclassification support vector machine achieving the requirements of thesubject in pipeline ultrasonic inspection signal automatic classification,many experiments showed that, this method is a simple and effectiveclassification algorithm.