A preliminary investigation into the automatic detection of diseased sheep organs using hyperspectral imaging sensors

Abstract

The post-mortem inspection process of livestock viscera at abattoirs is expensive and laborious, but it is essential for the detection and condemnation of defective edible organs and carcases due to food safety issues. Lesions in hearts, kidneys, livers, lungs, and their associated lymph nodes are amongst the most common offal defects found in abattoirs. Visible (VIS) and short-wave infrared (SWIR) hyperspectral imaging implemented in a multisensory platform were used to differentiate between sheep parenchymatous organs passed as fit (Healthy, n = 42) or not fit (Diseased, n = 47) for human consumption. Partial least squares discriminant analysis (PLS-DA) and random forest (RF) were used to classify organs as healthy or diseased in heart (n = 28), kidney (n = 15), liver (n = 24), and lung (n = 22). PLS-DA produced equal or greater classification accuracy and sensitivity than RF for all organs except for lung when the VIS sensor was used (means 84.4% and 78.3%, respectively). Livers and hearts (86.9%) showed higher accuracy than lungs and kidneys (75.9%). Limited differences occurred between VIS and SWIR sensors, although a single sensor tended to be more accurate than a combination of both. SWIR outperformed VIS in accuracy across all organs (84.8% vs. 76.3%), and the combination of VIS and SWIR was also accurate (83.0%). The use of hyperspectral imaging is an attractive proposition for the meat processing industry as a non-invasive imaging technology to detect defects in offal, and it can also provide automatic detection, saving time and labour costs.

Keywords

• Disease detection
• Automation
• Meat processing
• Sheep health
• Organ condemnation
• Inspection