<![CDATA[This research focuses on the development of a student facial classification model for attendance verification using Google Vertex AI AutoML. A total of 401 facial images representing 20 student classes were utilized, undergoing preprocessing steps including resizing to 224×224 RGB resolution and conversion to 8-bit format. Data augmentation techniques such as horizontal flipping, ±15° rotation, and brightness modulation were applied to enhance dataset variability. After refinement, 367 images were retained and divided into training (80%), validation (10%), and testing (10%) sets. The model was trained using the Edge TPU – Best Prediction mode in Vertex AI AutoML, resulting in an excellent performance with an average precision of 0.999, precision of 100%, and recall of 89.2%. The confusion matrix indicated that most classes were accurately identified with minimal recall errors. The finalized model was converted to TensorFlow Lite (TFLite) format and tested on edge devices, demonstrating efficient inference and accurate recognition. The findings affirm the effectiveness of integrating AutoML and TFLite to implement lightweight, resource-efficient face recognition systems suitable for student attendance applications on constrained hardware platforms.]]>
Copyrights © 2025
