Cross-correlation based Acoustic Signal Processing Technique and its Implementation on Marine Ecology

Abstract

As a scientific study of marine-life habitation, populations, and interactions among organisms and surrounding environment, marine ecology includes numerous fish and mammals as part and parcel. Marine fish and mammals have an enormous impact on marine ecosystems. Not only for their ecological values, but also for commercial purposes, a proper estimation of their population size is necessary. Besides, an efficient monitoring of populations and communities is the precondition of ecosystem-based management in marine areas. Most conventional techniques for estimating fish population are visual sampling techniques, environmental DNA (eDNA) technique, minnow traps, removal method of population estimation, echo integration techniques, etc., which are sometimes complex, costly, require human interaction, and harmful for inhabitation of marine species. In order to overcome these difficulties, an acoustic signal processing technique is proposed in this thesis. The method is based on a novel statistical signal processing technique called “cross-correlation” and different types of acoustic signals produced by diverse species of marine fish and mammals, like chirps, grunts, growls, clicks, etc. Our goal was to build a framework so that the technique can be implemented in practice. Therefore, we have investigated different tasks, which are crucial during its practical implementation like estimation with respect to different fish acoustics, different number of sensors and different distributions of fish and mammals. Similarly, we have carried an investigation to select the optimum estimation parameter for the technique. We have also analyzed different impacts, i.e., underwater bandwidth, SNR, etc., which have significant effects on practical estimation of this technique. From this research, we have found that chirp signals can produce better estimation results among the three fish acoustics, i.e., chirps, grunts, and growls signals. Among the three fish distributions, i.e., Exponential, Normal, and Rayleigh, Exponential distribution of fish and mammals produce better results. An increasing number of acoustic sensors provide better results in this technique. However, limited bandwidth of underwater channel poses a barrier during acquisition of fish signals, which has infinite bandwidth. To overcome this problem, a proper scaling is a mandatory task. We find that scaling factor 0.59512 for chirp signal and 0.55245 for grunt signal at 5 kHz underwater bandwidth. Similarly, a low signal to noise ratio (SNR) is also an impediment to obtain an accurate fish population. We have found that estimation with minimum SNR of 20 can perform like the noiseless estimation. These findings will immensely help the future researchers during practical implementation of the technique.