Title : Counting fish system using artificial intelligence
Abstract:
This article is about a counting fish system using artificial intelligence. We trained a neural network to count fry (fish in its young state) in controlled tanks. The neural network was developed using the YOLO (You only look once) architecture. It is able to identify fish and count them, this is done from images used as input to the network. Together with the neural network, with the addition of a graphic interface, the user can send photos taken with his smartphone and know the amount of fish he is raising in each tank, which gives him greater control of his production, whether in relation to the amount of food, medicine or data collection. The neural network can be used
to optimize the task of manually counting fish, a task that is subject to human errors. The counting made by people,
besides being slower, causes stress to the fish. When fish are stressed a lot, they end up developing problems in their growth and increasing the chances of diseases, which compromises production. This article will address important information about the neural network training process, such as: the importance of the quality of the input images, the functioning of the YOLO architecture (you only look once), evaluation metrics used, evaluation of the results obtained. Currently the network presents good results, reaching an effectiveness of 99% of hits. Initially, the neural network can identify and count fish in controlled tanks, but this could be the first step to take the idea further to natural environments, such as rivers, thus promoting an analysis of the local fish population in a given region.
What will audience learn from your presentation?
? Explain how we create the fishes dataset
? What is a neural net and how works
? Show the graphic interface and how the counting works
? Present the results
