A Large-Class K-shot Manchu Recognition Method for Based on N-ary ECOC

Manchu recognition is regarded as a K-shot learning problem due to the difficulty in obtaining training data. However, in practical applications, the number of categories that Manchu recognition is required to confront is extremely large, rendering the traditional K-shot learning algorithm inadequate. In this paper, a K-shot learning algorithm for large-category recognition is constructed. The fundamental strategy involves decomposing the original large-category classification problem into a series of smaller-category classification ones through N-ary ECOC (error correcting output coding, ECOC) technology before further processing. The algorithm consists of two stages: encoding and decoding. In the encoding stage, the large-category support set is decomposed into a series of smaller-category sub support sets by the N-ary ECOC coding matrix, and subsequently, multiple K-shot learning base classifiers are generated based on the sub support sets. In the decoding stage, the test sample is classified by the aforementioned base classifiers to form a prediction code. Eventually, the prediction code is compared with the coding matrix for error correction to determine the final class. The proposed algorithm attains a recognition accuracy of 87.8% on a 500-category Manchu dataset.