Development of Android Application using Seven Segment Display for Digit Recognition using Deep Learning

The proposed deep learning Android application performs digit recognition on seven-segment displays. It uses a CNN model trained on a custom dataset and optimized for ondevice inference using Tensor- Flow Lite (TFLite). The app will enable the user to choose a picture using the device camera or pick up a picture on the gallery. The prepro- cessing of the captured image optimizes the seven segment digits and the optimized image is sent to the trained CNN model for the classification. The identified numbers are presented on the interface of the application and they are stored on the device in SQLite database including times- tamps to enable users to access past records and analyze them. Data augmentation through rotation, scaling, and adjusting the brightness of the input image, the training of the model was performed in order to facilitate generalization. The precision, recall and F1-score are large in experimental evaluation confirming the accuracy and reliability of the of- freed approach. The lightweight aspect of the application makes it work in real-time on mobile devices with minimal resources, and the applications that could be fitted within could be automated meter reading and digital display monitoring.