Convolutional Neural Network based Marine Cetaceans Detection around the Swatch of No Ground in the Bay of Bengal

Abstract

The blue revolution of the blue economy (BE) on the way to build a golden Bangladesh is now the demand of the time. BE is a sea-based economy. The economy of exploiting the vast resources of the oceans and their bottoms. Which means that whatever is extracted from the sea, if it is added to the country's economy, it will fall into the category of BE. But the amount of resources of Bangladesh at Bay of Bengal (BoB) has not yet been surveyed properly. This paper deals with detecting marine cetaceans (MC) based on convolutional neural network (CNN) around the Swatch of No Ground (SoNG) in the Bay of Bengal (BoB). At first the possible MC living around BoB have listed for the training purpose of neural network (NN). Then the dataset (both training and validation or test) being trained to NN have created by extracting spectrogram images of the clicks, whistles or songs (CWS) of listed MC around the SoNG. Three types of test data (TD) such as original test data (OTD), synthetic test data (STD) and practical test data (PTD) have considered to validate the proposed method. The TD retrieved from the dataset is the original test data (OTD). The STD and PTD have derived from the OTD. Then the NN has trained with the training sets (TS) for the detection and classification of MC. After successfully completing the training process, the proposed NN has evaluated with three types of test sets and recorded the output to analyze the performance in detection and classification of MC. This model has successfully detected and classified the species of cetaceans with the accuracy of 100% for OTD, 88.88% for STD and 77.77% for PTD. The model has given wrong output of 3 and 6 false detection incidents for the STD and PTD respectively due to the underwater background sounds or ship sounds in the ocean. The method has simulated and validated using python programming language.