Aiming at the detection of loose parts in nuclear power plant under low signal-to-noise ratio (SNR) environment, an impulse response extraction method for loose parts based on blind deconvolution algorithm is proposed to reduce the false alarm rate and the false negative rate. Combined with Support Vector Machine classification identification algorithm, a detection method for loose parts of nuclear power plant is developed as well in low SNR environment. The simulatedstudies are carried out based on the test data from steel ball-flat plate drop experiment superimposed with real background noise, and the parameters involved in the blind deconvolution algorithm are optimized. The results show that the optimized blind deconvolution algorithm can effectively suppress the noise. Consequently, the impulse response signal with SNR as low as -20dB can be recovered well and the noise energy is reduced by 75%. The proposed detection method for loose parts has a very low false negative rate and a zero false alarm rate for noise and pulse interference even when the signal to noise ratio is as low as -14 dB. These results indicate the presented loose parts detection method has a good ability of anti-false positive and false negative.