An Aquaculture-Based Binary Classifier for Fish Detection using Multilayer Artificial Neural Network

Abstract

Fish detection, a specific task in computer vision system for fish monitoring, is challenging due to the complex characteristics of the captured images. A proposed approach in tackling this challenging task was to incorporate a multilayer artificial neural network to a computer vision system algorithm, implemented in aquaculture. This computer vision system algorithm captured the images from the aquaculture setup. Then, these captured images were processed. After that, the features out of these processed images were extracted and utilized to develop this multilayer artificial neural network. The best configuration, which is trained with the least learning time and tested with least mean square error and highest accuracy, was determined by adjusting the number of neurons in the two hidden layers. The multilayer artificial neural network with 50 neurons in the first hidden layer and 10 neurons in the second layer was considered the best configuration; it has achieved learning time of 3.374 ms, mean square error of 0.2315, and accuracy of 79.00%, hence, proving the competitiveness of this approach.