Desarrollo de un Dataset Especializado para la Detección Automática de Sacos de Balanceado de Camarón Aquaxcel GROWER 5 (25 kg), Aquaxcel Pre-Starter 2 (25 kg) y ECOFEED Purina (25 kg) utilizando el Modelo YOLOv8n

Automated processes in the logistics area are necessary to maintain efficiency and competitiveness at the operational level within a company, and the aquaculture industry is one of the sectors where this implementation is necessary. The objective of this work is to create a data set specialized in aquaculture products to be used with the YOLOv8n model for the automatic detection of aquaculture food packaged in woven polypropylene bags. Data collection was carried out in the parish of Hualtaco, El Oro, Ecuador, by the company ASOCAM, obtaining images of three types of bags, taking capture conditions such as lighting and different angles, ensuring their quality with inclusion and exclusion criteria, to be subsequently annotated with bounding boxes and divided into subsets for training, validation, and testing. The YOLOv8n model was trained in Google Colab, implementing improvements to balance the data classes, resulting in 2460 images for training, 307 for validation, and 308 for testing. The hyperparameters used for training were task (detect), mode (train), data (/content/data1/data.yaml), model (yolov8n.pt), imgsz (640), and batch (16). The dataset is trained with the YOLOv8n model, demonstrating an average accuracy (mAP) of 81.08% at the threshold (IoU) between 0.50 and 0.95, in addition to 89.02% for an IoU of 0.50 and an overall accuracy of 89.02%, which can be compared with studies similar to the project, such as employment in industrial environments. The overall error rate resulting from the evaluation is 0.0813, demonstrating acceptable performance for improving logistics processes with AI models as well as providing a basis for future research and applications within the industry. This project shows the construction of an image dataset for modernizing industrial processes with the YOLOv8n model for fast and accurate object detection. Keywords: Image data, Computer vision, Object detection, Convolutional neural network, YOLO.

• Prif Awdur: Manchay Montoya, Keyner Alexis (author)
• Fformat: bachelorThesis
• Iaith: spa
• Cyhoeddwyd: 2024
• Pynciau: DATOS DE IMÁGENES; VISIÓN POR COMPUTADOR; DETECCIÓN DE OBJETOS; RED NEURONAL CONVOLUCIONAL; YOLO
• Mynediad Ar-lein: https://dspace.unl.edu.ec/jspui/handle/123456789/30525