Tracking of Fish Behaviour Using Computer Vision and Deep Learning

Authors

  • Keerthi Samhitha B., Subhashini R.

Abstract

Deep Learning is expanding its application area and developing interest in processing techniques, visualization and feature extraction when compared to Machine Learning Algorithms. Deep learning methods have made it possible to create more efficient and sophisticated models of computer vision. The usage of computer vision applications is now becoming immensely important since these technologies advance. While using CV models to process the visual data sometimes it adds many layers to give expected output. To diminish this problem Neural Networks are used to progressively reduce the input data and calculate the most relevant data. With such rapid development of visual data processing techniques using Deep Learning it has become popular in many fields along with Aquaculture. This emerging techniques were applied on aquaculture and helped in developing the methods on fish farming, fish analysis, and detecting species of fish, or any aquatic animals and also used in fish behavior, water toxic analysis, etc. in this proposed system we use Deep Learning and Computer Vision techniques to predict fish behavior in different conditions like feeding, hypoxia, hypothermia, frightening, and normal. To obtain a data set of accurate precision we have a small setup a RGB camera to record the behaviors. All these data is processed with neural networks and computer vision techniques and compares the accuracy rate of each technique and noted as a part of experiment along with fish behavioral statistics under different conditions. To detect the fishes easily and concentrate on them we also used YOLO object detection algorithm. The results of precision, recall, specificity, and accuracy of the model on five different behaviours of fish are shown.

Downloads

Published

2022-08-14