<b>Real-Time Classroom Emotion Analysis Using Machine and Deep Learning for Enhanced Student Learning</b>

This research creates an innovative EfficientNet-B7-based Facial Expression Recognition model that delivers maximum accuracy performance for detecting emotions. Successful classification performance benefits substantially from EfficientNet-B7's application of compound scaling techniques which balances the entire network dimensions depth width and resolution. The characteristic distinctive to EfficientNet-B7 over standard architectural models involves its dual capability to perform accurate computations at reduced complexity levels. The model receives evaluation using KDEF at high-resolution as well as FER2013 at low-resolution through usage of SGD, Adam, and RMSprop optimizers. Experimental tests confirmed that EfficientNet-B7 operates with RMSprop optimizer to recognize emotions on KDEF at 91.78% accuracy superior to ResNet152's highest recorded accuracy of 88.77%. Performance levels declined to 57.56% on FER2013 because low-resolution images represent a great challenge to the model. Internal Batch Normalization (IBN) enters the model as an issue solution to halt gradient descent problems, which results in better model training stability and enhanced accuracy-loss patterns. The research demonstrates that FER performance benefits greatly when EfficientNet-B7 works in combination with IBN for high-resolution image processing. The research proves that EfficientNet-B7 stands as a reliable FER solution that shows potential usage in affective computing and human-computer interaction domain.<br>

本研究构建了一种基于EfficientNet-B7的创新型面部表情识别（Facial Expression Recognition）模型，可在情绪检测任务中实现最高精度性能。该模型分类性能的优异表现，很大程度上得益于EfficientNet-B7所采用的复合缩放技术——该技术可平衡网络的深度、宽度与分辨率等整体维度。相较于标准架构模型，EfficientNet-B7的独特之处在于其兼具双重能力：既能实现精准计算，又能降低复杂度。研究人员使用SGD、Adam和RMSprop优化器，分别在高分辨率数据集KDEF和低分辨率数据集FER2013上对该模型进行了评估。实验结果证实，当EfficientNet-B7搭配RMSprop优化器时，在KDEF数据集上的情绪识别准确率达到91.78%，优于ResNet152所记录的最高准确率（88.77%）。而在FER2013数据集上，模型性能降至57.56%，这是由于低分辨率图像对模型构成了较大挑战。研究人员引入内部批量归一化（Internal Batch Normalization, IBN）作为解决方案，以缓解梯度下降问题，从而提升模型训练的稳定性，并优化精度-损失曲线。研究表明，在高分辨率图像处理任务中，EfficientNet-B7与IBN结合使用可显著提升面部表情识别性能。本研究证实，EfficientNet-B7是一种可靠的面部表情识别解决方案，在情感计算（affective computing）和人机交互领域具有潜在应用价值。