Research on Intelligent Grading Method of Beef Marble Pattern Based on Few-shot Learning

Abstract

Abstract The beef quality grading model based on deep learning requires a large number of samples. Obtaining accurate beef samples requires manual classification or classification through more complex process methods, and the workload is huge. Aiming at the above problems, a beef marble pattern recognition model based on small sample learning is proposed. According to the national standard, a beef marble pattern grading data set was established by artificial classification method, and a lightweight CNN network was designed for image feature extraction. The pre-training of CNN was completed on the mini-ImageNet data set. The cross entropy loss function is updated with support set samples and the entropy regularization function is updated with query set samples to further optimize the parameter weights in the softmax classifier. Using cosine similarity to compare image feature vectors, softmax as a classifier to complete the task of image classification. The results show that the classification effect of this model is the best, and the highest accuracy of beef marble pattern recognition is 96.66 %. Under the premise of the same number of training samples, it is obviously better than other models.