A Short Video Classification and Retrieval Scheme Based on Transfer Learning

Short videos have become a new form of media, developing rapidly and showing great prospects. Given the exponential increase in the volume of short videos, manual labeling has become increasingly challenging and costly, rendering it feasible only for select short video classification tasks. These issues constrain the efficacy and efficiency of short video data applications at a large scale. To address this issue, this paper proposes to use the residual neural network model as the main deep learning model to extract and classify the features of short videos. To improve accuracy and minimize training time, a short video classification and retrieval system is developed using transfer learning-based model training, enabling the system to leverage ResNet’s deep network model to enhance classification performance while reducing training time and sample size, and achieved high accuracy on small sample datasets.