Intelligent video analysis platform for micro-classroom based on BiFPN-YOLOv5s algorithm

Abstract

As micro-teaching software advances, more and more teachers and students opt to use micro-classroom videos for learning. Traditional video analysis methods often perform poorly and have low accuracy when analyzing micro-classroom videos. To address this issue, the research introduces a novel approach for constructing an intelligent video analysis platform. This approach integrates a bidirectional feature pyramid network with the You Only Look Once version 5 small (YOLOv5s) algorithm. The bidirectional feature pyramid network classifies and integrates video image information. The research then utilizes the YOLOv5s algorithm to collect feature information from intelligent video images within micro-classrooms. The image feature information goes through an input-output process. Next, the research incorporates a fusion compression excitation attention module to improve the accuracy of model-based video analysis, ultimately building an intelligent video analysis platform for micro-classrooms. The results show that the proposed method achieves a recognition accuracy of 97.31% for various video image information, with a harmonic mean of 97.67%. Its average accuracy for video analysis reaches 0.91, significantly exceeding that of other algorithm models. In the practical effect analysis experiment of the video analysis platform, the micro-teaching software shows a memory occupancy rate of only 1.54% and a central processor occupancy rate of just 1.77%. Additionally, its response time is better than that of other video analysis platforms. These findings indicate that the proposed method provides high accuracy and strong practicality for intelligent video analysis in micro-classrooms.