Deboning represents a particularly labor-intensive operation in the poultry industry. Due to the natural deformation and variation of bird carcasses, automation of the cutting process has proved to be very challenging and has resulted in significant yield losses and bone chips. ATRP’s Intelligent Cutting and Deboning System uses 3D imaging and a robotic cutting arm to automatically perform precision cuts.

Cuts are focused on severing the tendons and joints on bird front-halves in preparation for the removal of the wings and breast meat. Recent efforts have focused on integrating three separate components (trajectory generation for each individual bird, bone detection, and force control) into a single functioning system.

Initial performance results were encouraging, and the team is currently refining the system by designing and fabricating an improved knife end-effector and expanding the use of force control. The team also plans to perform a more extensive statistical study of bird features and further explore active wing manipulation.

Screening deboned poultry product for bones is still an intensive manual process. In addition, estimating yield loss due to process inefficiencies is also very difficult to perform during production. ATRP’s Cone Line Screening System project team has developed a vision-based approach to address these issues.

Recent efforts have focused on developing and evaluating routines for performing bone detection and yield estimation on deboned poultry products. Researchers collected approximately 2,600 images of deboned product at a poultry processing plant. This included 2,500 images of product taken directly from the deboning line, and 100 samples of birds before and after performing a manual yield assessment process.

In addition, several hundred additional frames from different processing facilities were imaged in the laboratory. During testing the system was able to classify 100 per cent of missing clavicle bones from the test data. However, there was a high false positive rate of 20 per cent, primarily due to broken clavicles without missing bone chips.

Fan bone detection accuracy was 82 per cent. More promising was the yield results, giving a correlation of 90 per cent with the data tested from the field testing.

However, testing the yield estimation routines on birds from a different producer only yielded a correlation of 72 per cent. This approach still shows promise for detecting bones as well as demonstrating the ability of monitoring process yield in real time. Research is underway to refine the yield estimation and bone detection routines and perform robust field testing on the final pre-production prototype system.