SCAI: A Spectral Data Classification Framework with Adaptive Inference for Rapid and Portable Identification of Chinese Liquors

To achieve accurate, rapid, and portable identification of Chinese liquors based on spectral technology, this paper leverages the adaptive inference technique to solve the computational efficiency problems of existing deep learning models. We propose a Spectral data Classification framework with Adaptive Inference (SCAI), which leverages the Early-exit paradigm and can allocate appropriate computations for different samples on different portable devices. It can adjust the computation layers in each neural network block for different spectral curve positions of liquors so that it can focus more on important information. To our knowledge, this paper is the first attempt to leverage adaptive inference for liquor identification. The experimental results show that our method can achieve higher identification performance (+6%−+ 13% under the same budget) with less computational budget (1/6 for the same performance) than existing methods.