Embedded computer vision system designed to automate the classification of pig carcasses in slaughterhouses.
The Pig’Sel (Pork Image-Based Grading System for Estimating Lean) replaces or complements currently used optomechanical probes, such as the CGM.
This semi-automated device offers a non-invasive, contactless measurement method that minimizes the risk of cross-contamination while adapting to industrial processing speeds. The system is based on a compact architecture combining a Raspberry Pi microcontroller, a local display, and a 12-megapixel high-definition camera with autofocus. The software pipeline processes data in real time in less than one second.
After acquisition, algorithms correct optical distortions and perspective effects. An artificial intelligence model based on a U-Net convolutional neural network then performs anatomical segmentation of the carcass's fat, muscle, and bone structures. From the obtained masks, the program automatically extracts the fat (G3) and muscle (M3) thicknesses by calculating Euclidean distances. These values are integrated into the official regression equation to determine the Cut Muscle Percentage (TMP) with an accuracy of ±0.2 mm.
Pig’Sel also ensures complete digital traceability by archiving images, annotated masks, and metadata for quality control and carcass valuation.
The modular architecture allows for the integration of new indicators, namely the color of the fat and muscle, to define meat quality.
