Applying Imaging Technologies

Wednesday, January 17, 2024

In One Click: Imaging Technology for The Rapid Prediction of Pork Carcass Composition

By Argenis Rodas-González

Pork producers select pigs for slaughter based on the evaluation of live pig conformation and weight, which are labour-intensive and stressful for the pigs. However, most hog markets pay the producer based on pork carcass merit, which is determined by the percentage of carcass fat-free-lean muscle (FFLM).

FFLM is determined post-mortem using various carcass measurements, which are time-consuming, require skilled personnel, and use invasive and destructive techniques (i.e., ribbing or penetrating using an optical probe) and are therefore unsuitable for use on farms to select market animals. As a result, large numbers of pigs do not meet target specifications set by the packing plant leading to frequent penalization for animals that are too lean or over-fat.

In addition, since 2020, the Canadian swine industry has experienced unprecedented challenges caused by the COVID-19 pandemic affecting swine producers, including the inability to find a market for finished pigs and labour shortages. Currently, only limited tools are available to monitor the efficiency of growth, finishing and marketing of pigs in conventional production. The tools that are available require significant training, are time-consuming, and require significant animal handling that causes stress and endangers the safety of the animal and personnel.

Why apply imaging technologies?

Imaging technologies are non-invasive and non-destructive. These include 2D and 3D digital imaging techniques that accurately determine live weight, body score, animal welfare and health in the livestock sector. However, their performance depends on illumination conditions and depth distance (distance between the camera and the object).

In a preliminary study completed at the University of Manitoba, using a 2D camera and a limited number of growing pigs, researchers were able to predict live animal body measurements (length, width, height, and area of hams and loins) that were manually extracted from the images.

Using prediction equations, the researchers were able to determine, from moderate to high accuracy, live weight (46 per cent), ribeye area (75 per cent), backfat depth (76 per cent) and lean yield (74 per cent).

Preliminary data suggest that a multispectral camera can overcome the challenges faced in the 2D camera trial, including supplying adequate lighting and contrasting background (dark floor or wall). The application requires a more flexible and robust system for accurately determining carcass traits and FFLM prediction in pigs.

The multispectral camera consists of three sensors: Visible, thermal-IR (infrared) and ToF (time-of-flight). Integration of the sensor outputs provides a 3D multi-spectrum data framework in real-time. Traditionally, the multispectral image has been used in several fields, including the aerospace and pharmaceutical industries. To date, this approach has been used in the poultry industry to characterize several types of abnormal carcasses, including bruised, tumorous, and skin-torn carcasses.

Infrared thermography is another imaging method used in the swine industry. This method directly measures an animal's energy loss (i.e., heat transfer) by detecting body temperature changes. Infrared thermography can identify animal metabolic efficiency (feed efficiency), health status (i.e., inflammatory process, diseases, animal stress), reproduction (i.e., scrotal temperature) and determine methane production.

Getting the click

My team at the University of Manitoba includes graduate students Veronica Ndams and Ankita Saikia from the Department of Animal Science. We are working in collaboration with Alpha Phenomics and Animal Intrametric to provide solutions to swine farmers through the image technologies. The project is sponsored by Alberta Innovates and Results Driven Agriculture Research.

The project will test the multispectral camera and thermos camera systems to develop predictions of live weights and carcass traits and fat-free lean in pigs, based on the captured images and direct measurement of body composition (length, width, height, and area of hams and loins), weight data and post-mortem carcass data.

At the same time, within this framework, multispectral and thermal camera image results will be contrasted with other non-invasive and non-destructive technologies such as dual energy X-ray absorptiometry. This technology will be applied to the carcasses and primals to predict pork carcass traits and the quality and composition of primal cuts. The data will be used for future integration of precision technologies and may also assist in bridging the presently existing gap between producers and processors.

The goal of this proposed and unique research is to apply a modern artificial intelligence technique called convolutional neural networks to train machine learning models to extract and predict carcass parameters from live images directly. At the same time, the study will generate a report that farmers can use to manage their animals in real-time. The research offers an opportunity to create an autonomous monitoring system of grower-finisher pigs to reduce inefficiencies in swine production and decrease carcass nonconformities, providing economic and market advantages.

The proposed technology could reduce the dollar value loss for hog carcass nonconformities reported in the 2003 pork quality audit, averaging $8.08/carcass.

The largest losses, $1.32, were attributed to inconsistent live weight, carcass and wholesale cut weights. Thus, this technology could produce a uniform lot (i.e., decrease carcass nonconformities), increase swine producer profitability, and improve sustainability in swine production. BP