Measuring the weight of egg with image processing and ANFIS model
SEMCCO'11 Proceedings of the Second international conference on Swarm, Evolutionary, and Memetic Computing - Volume Part I
| Hi-index | 0.00 |
A blood spot detection neural network was trained,tested, and evaluated entirely on eggs with bloodspots and grade A eggs. The neural network couldaccurately distinguish between grade A eggs andblood spot eggs. However, when eggs with otherdefects were included in the sample, the accuracy ofthe neural network was reduced. The accuracy wasalso reduced when evaluating eggs from other poultryhouses. To minimize these sensitivities, eggs withcracks and dirt stains were included in the trainingdata as examples of eggs without blood spots. Thetraining data also combined eggs from differentsources. Similar inaccuracies were observed inneural networks for crack detection and dirt staindetection. New neural networks were developed forthese defects using the method applied for the bloodspot neural network development.The neural network model for blood spot detectionhad an average accuracy of 92.8%. The neuralnetwork model for dirt stained eggs had an averageaccuracy of 85.0%. The average accuracy of thecrack detection neural network was 87.8%. Theseaccuracy levels were sufficient to produce gradedsamples that would exceed the USDA requirements.