Target Material Identification And Classification Of The High Pressure Water-jet

In recent years the technology of high pressure water-jet has developed rapidly, and has been widely used in various industries. Its special advantages which in the process of machining, if we use the high pressure water-jet to detect mine, and to achieve the integration of mine-detecting and mine-sweeping, there is a good application value. However, the research on mines-detecting by high pressure water-jet technology, which makes mines detection and exclusion integrated is blank at present. Research on the detection of mines will be applied to high-pressure water jet technology proposed is an original research. To realize the function, the first step is to identifying the target material, one of the key technologies is reflected sound signals to identification and classification..The following researches are made in this paper on preprocessing of sound signals, extraction of eigenvalues and design of identification classifiers associated with landmine detection by means of target reflected sound:firstly, analyze transmission models of target reflected sound and background noise in project practices, and propose a target reflected sound signal preprocessing method of EMD threshold de-noising based on non-stability and other characteristics of target reflected sound signals; secondly, analyze foreign and domestic researches on eigenvalue extraction methods of sound signals as well as design methods of identification classifiers at present, state basic principles for target reflected sound eigenvalue extraction of wavelet packet short-time energy characteristic parameters and Mel frequency cepstral coefficients, and describe in details primary algorithms of multiple classifiers established by LS-SVM; thirdly, design test equipment and test plans, perform practical test for linear array high-pressure water jet detection targets and real-time collection of target reflected sound, and process reflected sound signals by above methods and test programs of target reflected sound signal de-noising preprocessing, eigenvalue extraction and multi-class classifiers complied based on MATLAB; at the same time, make analysis and contrast test of influence factors of target material identification, and use test results to analyze relevant effects of target internal structure and outline size on the target material identification.The experimental results showed that:according to the contrastive analysis, both signal to noise ratio (SNR) and relevant coefficients after EMD threshold de-noising are higher than that of wavelet threshold de-noising. Besides, detailed information of original signals is properly retained. Therefore, EMD threshold de-noising is used in this paper. In this paper, the two eigenvalue (i.e. wavelet packet short-time energy and Mel frequency cepstral coefficient) extraction methods with different features. In the target material identification method established by combining wavelet packet short-time energy features and LS-SVM, the rate of identification is97.5%, and the total time is3.023815s while selecting radial kernel function; while in the target material identification method established by combining Mel frequency cepstral coefficient and LS-SVM, the rate of identification is95.42%, and the total time is1.902854s while selecting multinomial kernel function. To establish LS-SVM target material classifiers of parameter optimization by using cross validation and grid search in combination should improve the recognition rate of6.25%.Upon overall consideration of effects of internal structure and outline size on target material identification results and based on the results of influence factor analysis test, the maximum influence of internal structure and outline size on target material identification is3.125%and8.4375%respectively, which is relatively small.Consider recognition accuracy and system real-time factors, choose the wavelet packet short-term energy and combination of LS-SVM. Taking these research results can be obtained:this design algorithm can effectively achieve the target object material identification and classification.