Target identification method based on continuous spectral characteristics of Gammatone frequency band

Abstract

The invention discloses a target identification method based on continuous spectral characteristics of a Gammatone frequency band. The target identification method comprises the following steps: firstly, performing windowing treatment on original target radiation noise data, selecting a Hamming window, establishing a corresponding window function, and performing fast Fourier transform on a windowed signal; determining number of Gammatone filter banks, determining center frequency of each filter at equal distance for an original signal frequency band, calculating impulse response of the Gammatone filter banks, performing fast Fourier transform on the impulse response, performing normalized processing, and establishing a corresponding Gammatone filter bank impulse response function. Comparedwith a conventional continuous spectral characteristic extraction, classification and identification method, the target identification method firstly performs preliminary identification, performs continuous spectrum in a sub band, can extract a stable sub band continuous spectrum as a typical sub band sample, and is more precise in whole frequency band continuous spectrum estimation, so that a correct identification rate of target radiation noise identification is increased.

Description

technical field [0001] The present invention relates to a target recognition method of sub-band continuum features, in particular to a target recognition method based on Gammatone sub-band continuum features, which is mainly applied to feature extraction and classification recognition of surface / underwater target radiation noise The invention belongs to the technical field of underwater acoustic signal processing. Background technique [0002] At present, the target feature extraction and classification recognition based on the target radiation noise data detected by passive sonar has been widely used in the military field. Since the target radiation noise and environmental noise are both random signals, and the environmental noise has obvious time-varying characteristics, Therefore, it is difficult for a specific method to realize the classification and recognition of multiple types of similar objects. [0003] Generally speaking, the Gammatone filter is a standard cochlea...

AI Technical Summary

Problems solved by technology

[0002] At present, the target feature extraction and classification recognition based on the target radiation noise data detected by passive sonar has been widely used in the military field. Since the target radiation noise and environmental noise are both random signals, and the environmental noise has obvious time-varying characteristics, Therefore, it is difficult for a specific method to realize the classification and recognition of multiple similar objects.

[0003] Generally speaking, the Gammatone filter is a standard cochlear auditory filter. The Gammatone feature extraction method is a typical auditory feature extraction method. This method divides the target radiation noise into subbands by simulating the human ear, and then performs energy accumulation and subsequent transformation. , to achieve feature extraction of target radiation noise, for targets with similar energy in sub-bands, it will lead to similar target auditory features, low target recognition accuracy, etc.

[0004] The continuum is a polynomial fitting of the power spectrum of the detection noise, and the sub-bands are equally spaced for energy accumulation as a feature vector. However, some sub-bands have high noise energy and low signal energy, resulting in a decline in the accuracy of recognition.

Images