ℓ0−normℓ0−norm regularized minimum entropy deconvolution for ultrasonic NDT & E

Ultrasonic backscattered echoes can be modeled as a convolution process. A broadband pulse is modulated by a band-limited system, which distort the measured signals. Consequently, deconvolution techniques have been widely used to improve resolution and quality of ultrasonic signals. Without the prior assumption of distortion function, high-order statistics methods are wildly used; one of the optimality criteria is MED (minimum entropy deconvolution). This paper describes an improved approach, which uses the ℓ0−normℓ0−norm regularization algorithm with the minimum entropy based deconvolution, which gives the benefits of faster convergence of algorithms and increase robustness to additive noise and inverse filter length. The ℓ0−normℓ0−norm regularization create a sparsity representation in transform domain, which sparser than time domain can acquire high resolution compared to conventional MED. The kurtosis is used as iterative stop criterion to indicate the minimization of entropy. Moreover, ℓ0−normℓ0−norm regularization can also eliminate the draw back of slow iteration caused by long filter. Simulated results show the efficiency of the algorithms improved in this paper and prove the parameters relationship proposed. Finally the experimental verification is performed on two different specimens with good result obtained. The result show that the methods designed in this paper is more suitable in ultrasonic NDE.