Curriculum classification network based on margin balancing multi-loss and ensemble learning

In this paper, we propose a curriculum classification network based on margin-balancing multi-loss and ensemble learning method, which solves the problems of difficult classification in different age groups and uneven distance within ages groups on the basis of the orthopantomograms. Firstly, due to the characteristics of teeth development, a curriculum learning framework is proposed and divides the sample easily into high-age and low-age groups, which reduces the error of dividing low-age into high-age groups or vice versa. Secondly, owing to minor differences of teeth development in the high-age group, the margin-balancing multi-loss is proposed to optimize the classification interface. Thirdly, considering the distinguishable characteristics in the low-age group and large range and non-concentrated features, an ensemble feature learning method is proposed to improve the classification reliability. Extensive experimental results on the orthopantomograms dataset demonstrate that the proposed method can perform well on different networks and accomplish precise age estimation.(c) 2023 Elsevier B.V. All rights reserved.

Automated summary

In this paper, a curriculum classification network based on margin-balancing multi-loss and ensemble learning method is proposed to address the difficulties of classification in different age groups and uneven distances within age groups using orthopantomogram images. The proposed method includes a curriculum learning framework, a margin-balancing multi-loss, and an ensemble feature learning method to improve classification accuracy. Experimental results on orthopantomograms dataset demonstrate that the proposed method performs well on different networks and achieves precise age estimation.