HamCapture

Coordinator: Università degli Studi di Padova

Scientific Officer: Giuseppina Schiavo

Duration: 12/10/2023 - 12/10/2025

Research group: Giuseppina Schiavo, Samuele Bovo, Mohamad Ballan.

The availability of objective, reliable and consistent quality evaluation methods is of outmost importance in the PDO dry-cured ham sector, which success is based on globally recognizable and consistent quality characteristics of the hams. The project aims at developing an automatic assessment system, based on 2D-3D image analysis coupled with hyperspectral imaging (HSI), exploiting innovative multivariate machine learning methods applied to classification and regression problems, to provide consistent and reliable automatic classification of green hams for visual defects and other characteristics of interest that can potentially affect the dry-curing process and the weight losses during dry-curing. The project aims also at expanding the knowledge related to the biological processes driving ham quality by identifying differentially expressed genes associated to ham quality traits. The identification of differentially expressed genes may provide insight into molecular events occurring in early postmortem muscle and likely influencing the metabolic and biochemical processes during the conversion of muscle to meat in hams characterized by different aptitude to dry-curing. Images will be captured from 1,000 hams which will also be scored by multiple specialists in order to obtain a common classification for visual defects, and measures of roundness, fat thickness, color of the exposed muscular tissue. Approximately 500 hams will be analyzed for lean meat and subcutaneous fat composition. Meaningful features will be extracted from the 2D-3D images and used to train models for the prediction of quality traits. The data obtained with HSI will undergo chemometric analyses to predict lean meat and fat composition. A number of multivariate methods for classification like partial least square regression-discriminant analysis and machine learning techniques (e.g. convolutional neural networks, random forests, support vector machine, deep learning) will be tested for ham classification. Multivariate regression techniques (e.g., partial least square regression, LASSO, and machine learning methods) will be used to obtain predictions of ham weight, fat thickness and chemical composition from images. 500 hams will be selected for high-throughput genotyping analyses and a smaller subset of 60 hams will undergo RNA sequencing. Genotyping will enable to perform association analyses with one or more phenotypes related to meat composition, visual defects and aptitude to dry-curing. RNA sequencing will enable to identify differences in the gene expression for the genomic regions that are responsible of the regulatory processes that affect tissues structure and are involved in meat and fat composition. The project offers a unique opportunity for the ham industry to enable process monitoring while at the same time generate highthroughput phenotypes and gain new knowledge to be incorporated in genetic evaluation procedures of pig lines used for dry-cured ham production.