Chicken egg quality prediction using ovoscopy, depth sensing, computed tomography, computer vision and machine learning

• Grant number: 22/07446-3
• Support Opportunities: Regular Research Grants
• Start date: November 01, 2022
• End date: October 31, 2025
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Agreement: FACEPE
• Principal Investigator: Denis Henrique Pinheiro Salvadeo
• Grantee: Denis Henrique Pinheiro Salvadeo
• Host Institution: Instituto de Geociências e Ciências Exatas (IGCE). Universidade Estadual Paulista (UNESP). Campus de Rio Claro. Rio Claro , SP, Brazil
• Associated scholarship(s): 24/02575-5 - Extraction of egg morphometric measurements from Computed Tomography images using Computer Vision, BP.TT; 23/06422-6 - Extraction of egg morphometric measurements from Computed Tomography images using Computer Vision, BP.TT; 22/15667-0 - Detection, segmentation and classification of eggs in ovoscopy images, BP.TT; 22/14804-3 - Extraction of egg morphometric measurements from computed tomography images using computer vision, BP.TT

Abstract

Artificial intelligence applied to machine learning can be used to optimize the evaluation of consumer eggs. The objective of this project will be to predict egg quality based on attributes related to eggshell and freshness using measurements obtained by candling, depth sensing and computed tomography, associated with computer vision and machine learning techniques. A total of 1,000 white eggs will be used with 0 (fresh), 7, 14, 21 and 28 days of storage, with 200 eggs in each period. The images will be obtained by candling, depth sensing and computed tomography and will be annotated. The same eggs will be subjected to destructive evaluation techniques to generate morphometric reference values for training artificial intelligence models and validation of results. Through the images obtained, computer vision techniques (Mask R-CNN, Fully Convolutional Network, ConvNeXt, KPConv, Ellipse R-CNN and others) will be used to obtain attributes related to egg quality. In order to define the quality and the recommendation for the destination of the eggs, the data collected from the images (automatically) and from the destructive techniques (manually) will be submitted to pre-processing (data cleaning and selection of features), model construction (regression logistics, random forest, artificial neural networks, deep neural network architectures, explainable AI models and techniques). The tested models will be evaluated by accuracy, F-1 score, AUC, Matthews correlation coefficient, Kappa score and log loss. It is expected that this investigation will contribute to the advancing the state of the art of automatic evaluation of egg quality. (AU)