A yolo-based student state detection method

Abstract

The invention discloses a student state detection algorithm based on YOLO, comprising the following steps: S1: improvement of YOLO; S2: adding bottleneck attention model BAM after the original DBL component, and then adding convolution after passing through two DBL components Attention model CBAM; S3: The modified network is trained, and the present invention relates to the target detection technology based on YOLO. The present invention adds a bottleneck attention model (BAM) and a convolution attention model (CBAM) on the basis of the YOLO network, and improves the precision of the YOLO network while ensuring a relatively high detection rate. At the same time, we apply it in the teaching classroom to realize the detection of students' listening status, which is convenient for teachers to understand and manage the classroom situation. The YOLO network with the attention mechanism has been added. After testing on the VOC 2012 data set, the speed can reach On the own student data set, the accuracy rate has been improved compared with the original network.

Description

technical field [0001] The invention belongs to the technical direction of target detection in the field of computer vision, and specifically relates to a method for detecting student status based on YOLO. Background technique [0002] Object detection is an important branch in the field of image processing, that is, to frame the range and category of objects of interest in an image. At present, target detection is divided into two genres: "two-stage" and "one-stage". The former is based on the idea of ​​locating candidate regions first and then classifying them, represented by the RCNN series, and the latter is directly performing candidate box regression and classification. , represented by YOLO, SSD, etc. YOLO is the fastest detection rate among them, but the detection accuracy is not enough. [0003] Among the technologies closest to the detection of student status, the paper "Research on the Algorithm of Student Classroom Behavior Detection Based on Faster R-CNN" uses...

AI Technical Summary

Problems solved by technology

[0004] The disadvantages can be high cost, low efficiency, time-consuming and similar problems. Due to the limitation of the Faster R-CNN network itself, the detection rate is very low, and the characteristics of the student classroom behavior data set itself (surveillance video images are different from general video images , it cannot obtain the positive information of the target, and there is a certain angle difference), so although the original paper can realize the detection of classroom behavior, it is lacking in real-time and high efficiency. If it is applied in the actual scene, it will be there are some problems

Images