Online Non-Contact Weighing Duck Carcasses Using 2d Images and Deep Learning

In the poultry meat processing industry, non-contact weighing methods based on computer vision have become increasingly popular due to their low costs and high efficiency. However, the accuracy of the traditional 2D image weighing techniques is reduced by the overlap and sway of carcasses. In this paper, we propose a more accurate regression model, Vggx, to weigh duck carcasses through image processing and deep learning. We adopt an image segmentation technique to obtain 2D images of a single duck section under a complex background. The images of the same duck are fused together to estimate its 3D information. Then, we use a convolutional neural network regression model to estimate the weight of duck carcass with the acquired 3D information. The Mean Absolute Percentage Error of our model is 1.46% and the average weight error of our model is 23.1 gram. Experimental results demonstrate that our method is high performance.