Use of Deep Learning for the Segmentation of Aquaculture Fishponds in the State of Minas Gerais, Brazil

Aquaculture has emerged as one of the fastest-growing food production sectors in recent years, playing a pivotal role in global food supply. This study explores a novel approach to satellite remote sensing, utilizing the Google Maps Static API in conjunction with a YOLO-based segmentation model, to map aquaculture fishponds in Minas Gerais, Brazil. Despite challenges such as shadows, vegetation, and some image limitations about the API, the method successfully segmented aquaculture fishponds, achieving an accuracy rate of 97.6% in estimating fishpond areas and a mAP50 of approximately 91.14%. The developed tool provides valuable support for aquaculture monitoring in Minas Gerais. Future research aims to refine the segmentation model, compare its effectiveness with expert identifications to identify new aquaculture locations across the state of Minas Gerais and address challenges related to isolated bodies of water, among others.