Segmentação de instâncias para estimação do comprimento de peixes utilizando técnicas de inteligência artificial

Abstract

In recent years, there has been a growing interest and concern for good practices and animal welfare in livestock farming. This is no different in fish farming, where fishes are sentient beings and respond rapidly to harmful stimuli, which can lead to direct and indirect negative effects on production. Therefore, proper management with low stress-inducing conditions can have a positive impact on production. In this context, the objective of this study was to develop an efficient alternative method for estimating fish length. To achieve this, the study employed Artificial Intelligence (AI) techniques as an alternative to traditional management methods that subject the animals to environmental stressors. The study initially collected 679 images of fishes at different developmental stages. Subsequently, an instance segmentation model was trained using the YOLOv8 architecture. With the segmentation model in place, a code was developed to estimate the length of each fish in the image based on the masks generated by the segmentation. This code was implemented in the Streamlit platform, resulting in an interactive and user-friendly web application. The application allows users to upload their images and obtain accurate results regarding the quantity of fishes in the tank and their average length. Additionally, it offers visualization features, such as displaying the segmentation and the length of the fishes, along with a line connecting the farthest points. The analysis of the results revealed an accuracy of approximately 88.2% concerning the actual sizes of the fishes. The study also found that light reflection on the water and shading during image acquisition negatively impact the results, emphasizing the importance of considering these factors and seeking ways to reduce them to achieve an even more precise length estimation. Based on the obtained results, it can be verified that the developed model has the potential to replace traditional management, ensuring better practices of animal welfare. This work also provides a solid foundation for further research and development in the field, propelling fish farming to reach new levels of excellence.