Desarrollo de un módulo de visión computacional para el conteo y detección de camarones mediante una aplicación móvil

Abstract

This project develops a computer vision module integrated into a mobile application for the automatic counting of shrimp (Litopenaeus vannamei), which is the first step in implementing biomass estimation in ponds. The proposal responds to the need to optimize sampling and monitoring processes in aquaculture, reducing stress on organisms and improving accuracy compared to traditional manual methods. The methodology included the capture and curation of a specialized dataset, the training of a model based on the U-Net architecture for shrimp detection, and the implementation of post- processing techniques to separate individuals and count them. The pipeline, which includes image capture and analysis by U-Net, was integrated into the existing mobile application, incorporating assisted capture, reporting, and data history management functionalities. The results show that the system can perform automatic counts with an average error of 3.35 individuals in low-complexity images, although limitations persist in high-density or overlapping scenarios. Even so, the technical feasibility of field integration and the potential for improvement through the use of larger datasets and advanced segmentation and counting models were verified. Keywords: Smart aquaculture, Automatic counting, Biomass estimation, Litopenaeus vannamei dataset.