Deep Analysis on Fisheries Classification

Classifying and detecting fishes in the wild is still an issue because of the various poses of fishes, ungovernable environment factors and the restricted quantity of data for training the model. The paper proposes an analysis of the techniques to classify different types of fishes caught on fishing boats without interference from humans using a Convolutional Neural Net (CNN). Different optimizers and different learning rates have been used and their findings have been evaluated in order to see how the two distinct CNN models work. To produce a comparative analysis of how one stacks against another in this particular problem, a dense CNN model, CNN 1 and a sparse CNN model, CNN 2 are built and analysed. The dataset used is publicly available on "The Nature Conservancy Fisheries Monitoring". Among the other models analysed with RMSprop and Adam optimizer, the dense CNN model with Stochastic Gradient Descent (SGD) optimizer offers the best outcome. A multi logarithmic score of 1.68087 is given.