The multi-mode operation decision of cleaning robot based on curriculum learning strategy and feedback network

In this paper, the model and its curriculum learning method of garbage hierarchical classification and corresponding operation mode decision in home environment are proposed from the perspective of cleaning robot. In order to realize the hierarchical learning of garbage attribute concept, this paper designs a learning model with iterative feedback network as the backbone network. In the early stage of iteration, the model focuses on learning the state of garbage, in the middle stage, it focuses on the appearance attributes of garbage, and the specific categories of garbage in the later stage. At the same time, the attention module is introduced to achieve different levels of feature expression learning, which further improves the performance of the model. The evaluation was conducted on the collected garbage data set and the public CIFAR-100 and Stanford Cars data sets, which verified the effectiveness and wide applicability of the proposed method.