Carcass image segmentation using CNN-based methods

Abstract

Carcass grading can be used as an important metric to determine meat quality. However, carcass grading is usually performed by a specialist, making it a subjective and error-prone task. To increase the accuracy of such task, image-based systems have been proposed in the literature. One of the most important parts of an image-based system is the image segmentation, which aims to identify the regions of the carcass in the image. In this paper, we propose the use of two recent image segmentation methods called Superpixel + CNN (Convolutional Neural Network) and SegNet. To evaluate both methods, we have also built a dataset of carcass images and their ground-truths. Results of approximately 96% of pixel accuracy show the robustness of the methods in carcass image segmentation. The novelty of this work is the proposal and comparison of recent deep learning methods that use CNN and superpixels in carcass segmentation. In this way, the methods can be used in carcass grading systems to increase the accuracy of the grading task.