Optimized Matching Pursuit For Ecological Environmental Sounds Recognition

In recent years, the protection of ecological environment has drawn wide public attention. Ecological sounds recognition is to extract features from natural ecological sound signals and then make recognition, which has a great meaning in protecting environment and knowing current ecological condition. In real situation, the background noise is complex and ever-changing, seriously degrades the recognition accuracy. Therefore, in this paper, we propose an optimized matching pursuit (MP) signal decomposition method and a new multi-band signal reconstruction for reconstructing the ecological sounds and have a focused research on robust anti-noise ecological sounds recognition (ESR) framework. The main work includes the following:1) Optimized signal sparse decomposition. Aimed at the high computation complexity of signal decomposition, we proposed optimized orthogonal matching pursuit (OMP) based on GSO optimization. Compared with MP, the improvement of OMP lies in its accelerating of convergence, and the enormous computation is the common weakness of this two algorithms. We adopt GSO to optimize the process of selecting best atom in the dictionary. Thus, decompose the sound signal rapidly.2) Anti-noise signal reconstruction. Aimed at sparse signal reconstruction unable to eliminate noise effectively, we proposed the OMP based two-stage multi-band signal reconstruction. Firstly, the OMP is employed to sparsely decompose the original signal, thus the high correlation components are retained to reconstruct in the first stage. Then, according to the frequency distribution of both foreground sound and background noise, the signal can be compensated by the residual components in the second stage. Via the two-stage reconstruction, highly non-stationary noises are effectively reduced, and the reconstruction precision of foreground sound is improved.3) Composite feature extraction and classification framework using deep belief networks (DBN).According to the analysis of ecological sounds, we employ DBN to model the composite feature sets extracted from reconstructed signal. This composite feature sets retaining rich temporal-spectral information, and make up the defects of the MFCC which is sensitive to noise.In this paper, ecological sounds include bird sounds、animal sounds and insect sounds.60 subclasses of clean samples are mixed with various environmental sounds to simulate the real situation. The two-stage approach is employed to reconstruct all these sounds to reduce the noise. And then, comparison experiments in different scenes under different SNRs are constructed using DBN based ESR framework to make identification. The experimental results show that the combination of OMP and multi-band reconstruction can effectively suppress the noises. Thereby improve noise immunity and robustness of ESR system. Compared with the existing methods, the framework this paper proposes achieved superior recognition performance in different environments under different SNR, which is especially suited in low SNR circumstances.