Detection of pest infestation by preprocessing sound using vector quantization

In the recent past, economic damage due to infestation of pest on palms could be mitigated significantly by early detection and treatment. Tests were conducted with currently available acoustic instrumentation and software to assess the sound impulses produced by larva's locomotory and feeding activities. The incorporation of bursts into analysis significantly assisted in applications where consistent activity patterns of hidden pest could be identified. The purpose of this research is to realize the effectiveness of a text independent identification system making use of cepstral coefficients and vector quantization. The identification system will be making use of MFCC. The MFCC extracted was then matched to all available sound codebooks that have been stored. The codebook that returns the lowest quantization error will belong to sound contained in audio input file. This confirms the result of detection of the particular species of interest. The data resulting from the analysis will serve as key characteristic in identifying presence of the pest species to which sound belongs.