Fast classification of indoor functional areas based on a pre-trained lightweight network

Wenbai Chen, Guohong Yi, Huixiang Liu, Qili Chen, Weizhao Chen, Peiliang Wu

Abstract: Lightweight networks can provide a solution to the high equipment requirements of traditional classified networks and the popularity of mobile and embedded devices. This work uses the trans-fer learning method to train MobileNetV3 to obtain a lightweight, fast, and accurate indoor function area scene classifier. The network is first pre-trained on a large dataset to obtain its weight, and then the pre-trained weight is transferred to an indoor functional area dataset for training using a feature extractor. Finally, a high-performance indoor scene functional area classifier with high accuracy, short response time, and low memory requirements is obtained.