LHNet: Logic-assisted Hierarchical Neural Network Model for Image Classification
Shuchang Ye · 2023
Neural networks cannot handle logic, but logic is an efficient way to perform classification when dealing with downstream tasks. In this study, domain knowledge is applied to set up logic for traffic sign classification. According to the similarities and functionalities of traffic signs, the dataset is divided into 4 subsets: prohibitory, danger, mandatory, and others. The heavy multi-class image classification task is split into the integral of several light multi-class image classifications, which is connected via logic justification. The complexity of model is significantly reduced and the inference time is much quicker.