In polyethylene pipe manufacturing processes, the precise measurement of the pipe thickness is an important technical problem, which affects the quality of the pipes produced. In the terahertz non-metal thickness transmission measurement process, the sample thickness parameter is obtained by measuring the optical path time difference of the two transmitted signals and material refractive index. To accurately extract the optical path time difference of the two transmitted signals, signal characterization of the original time domain signal is required, which is a mature and effective technical means to characterize the signals impulse response functions. An impulse response function can be obtained using a deconvolution technique. However, when the echo signal is weak or the system signal-to-noise ratio is low, the interference signal increases, causing distortion and flooding of the signal, and thus, the signal cannot be accurately extracted. In this study, we first pre-process the time domain signal with a Butterworth filter and perform an improved self-precision analysis on the processed signal. Second, we perform Gaussian deconvolution processing for autocorrelation signals. Finally, we obtain an improved impulse response function. The improved method aims to solve the problem of signal characterization in the field of terahertz non-metal measurement. The improved algorithm enhances the clarity of the signal and solves the distortion or flooding of the impulse response signal caused by the interference of the clutter signal. Consequently, the signal-to-noise ratio of the impulse response signal is improved.