Research And Implementation Of The Key Technology Of Students' Fatigue State Detection Based On CNN

As the main body of teaching activity, learners play an important role in it. Learning state as the reflection of the situation of students' physical, psychological and other various aspects, has become a hotspot in the field of education research. While as one aspect of learning state, fatigue state is better reflect the students' performance in class, studying it is helpful to understand students and teaching evaluation. Therefore, the study of students' fatigue state is necessary.Facial features are the effective characterizations of students' fatigue state, while researches have proved that eyes are more closely associated with fatigue state. The current detection process generally includes face detection, eye location and detection, feature extraction, eyes' and fatigue state evaluation. Among them, face and eye detection and the judgment of eyes' state are the core problems. But the main problems of existing research are:eyes positioning is vulnerable influenced by the external environment; state judgment need to define features and extract them. Only these key problems have being solved, can the judgment of students' fatigue state will be effectively achieved.Aimed at the problems existing in the traditional method, in this paper, a kind of students' fatigue state judgment method which is combining face detection with convolutional neural networks technology is put forward. The method can overcome the interference of light, occlusion, angle and so on environmental factors on eyes to a certain extent and can avoid the artificial features extraction operation. The main contributions in this paper are as follows:1. A method combining face detection with convolutional neural networks technology is proposed and the key technology is described in detail. Because of the complexity of convolutional neural networks, the influence of base_lr, the layer number of convolution and other parameters on the model is explored through a large number of experiments.2. In this paper, AdaBoost face detection based on Haar features is realized, and then the sample database for convolutional neural networks is build. By analyzing a large number of experiments, the suitable network structure for the database is confirmed and the classification model for eyes' state is got.3. In order to improve the classification accuracy of the model, the model is optimized through the expansion of data set and the relevant parameters'adjustment. The model implemented the students' fatigue state judgment. Results of large number experiments and comparison with traditional implementation method show that the optimized model has good classification ability to the state of human eyes, and system avoid a large amount of image processing and feature extraction operation.