Solution based on convolutional neural networks for automatic counting of aquatic animals

Aquaculture is the process of cultivating organisms with a predominantly aquatic habitat, being today an important activity in human food production. Despite its importance, there are still several activities that are carried out almost exclusively manually. Among them, it is possible to highlight the counting of animals. Therefore, this study presents a set of solutions to support the activity of counting aquatic animals using computer vision and machine learning techniques, through deep learning, with the differential that the end user will be able to access the solutions via a smartphone. Currently, the model is 99% accurate. A counting model based on YOLOv4 was also developed, which reached 98.50% of mAP and 98.70% of accuracy, thus obtaining an excellent result.