Multi-task convolutional neural networks (MTCNN) have garnered substantial attention for their impressive feats in image classification. However, MTCNNs tailored for the dual tasks of face recognition and facial expression recognition remain relatively scarce. These networks often entail complexity and substantial computational demands, rendering them unsuitable for practical systems constrained by computational limitations. Addressing this gap, the present paper proposes an efficient MTCNN model that capitalizes on cutting-edge architecturesresidual networks and dense-connected networks-to adeptly execute face recognition and facial expression recognition tasks. The model's efficacy is rigorously assessed across wellknown datasets (JAFFE, CK+, OuluCASIA, KDEF) as well as collected images from learners (HOUS22). The proposed model consistently attains high accuracy levels across all datasets, with a notable minimum accuracy of 99.55% on testing data. These outcomes stand as a testament to the model's remarkable performance, particularly in relation to its streamlined design. Moreover, the model is seamlessly integrated with an online learning management system, furnishing a versatile means of monitoring learning activities to enhance the overall training quality.