This study proposes a face detection algorithm based on a convolutional neural network considering the scenario of a classroom, where the faces of students sitting in the back rows might not be visible. First, the algorithm extracts face features in two stages using a residual neural network. Then, it builds a feature pyramid and combines the Softmax loss function with center loss function to train a face recognition model based on a proper activation function. Upon applying the algorithm to the Wider Face dataset, it achieves an accuracy of 95.2% and mean average precision values of 93.0%, 87.3%, and 58.3% for three levels of validation sets, respectively.