In response to the fundamental need for accurate recognition of tense faces wearing masks, this paper presents an innovative approach that takes advantage of parallel two-stage deep learning methods combined with hybrid meta-heuristic algorithms. The challenges associated with mask-wearing detection are addressed through a comprehensive framework that leverages state-of-the-art technologies and diverse inputs. This method includes a parallel algorithmic strategy, which optimizes the detection of faces with and without masks for accuracy. A special algorithm is used when detecting unmasked faces, while detecting masked faces uses a separate algorithm. In addition, an integration of multiple data sources including masked face images and inputs from temperature sensors increases the recognition accuracy. The main focus of this research is on data clustering, where datasets are organized based on their volume, then classification is performed using a proposed convolutional neural network. Duplicate features are carefully removed from each cluster, and parallel post-processing is performed by differentiating algorithms. In this study, two hybrid algorithms are introduced, and with the increase in data volume, additional algorithms can be easily inserted to provide scalability and increase accuracy. This innovative approach demonstrates the ability to significantly improve the accuracy and efficiency of complex facial recognition systems and addresses an important need in the fields of security to public health and beyond. Also, with the advancement of technology and the progress of research in this field, the possibility of improving the accuracy of detecting contracted faces is still promising.