Facial expression recognition by computer vision is an important task with many possible uses. To improve the classification effect, we developed and created a lightweight convolutional neural network (CNN) in this research that can detect face expressions both in real-time and in bulk. We build a real-time vision system to test the effectiveness of our approach. Automated facial impression recognition with real-time accurate interpretation is one of the assistive technologies that can be used to address the problem mentioned previously. With this method, seven facial expressions can be clearly identified and communicated to the relevant parties. This model employs a partial transfer of learning methodology, employing a CNN that has been specially trained to recognise face emotions. A new model is developed that can transfer characteristics between datasets. Making use of the freshly trained CNN at the core of the suggested method is a lightweight, portable facial expression recognition system that has excellent detection accuracy and wireless connection. The model also employs global average pooling instead of traditional fully connected layers. Finally, the FER-2013 dataset is used to test our model. Merely 2.9% of the 16 GB of RAM, or 0.396 GB memory, are required to do the face expression classification assignment. The weight of our model can be stored in 872.9 kilobytes files and has achieved 85% accuracy on the FER-2013 dataset. Additionally, it has strong recognition and detection capabilities for figures that are not included in the collection.