Research: CNN Model Comparision For ANPR (Indian Numberplates).

Includes Models like VGG-9, MobileNetV2 and Le-Net for Classification of Numberplate Digits and Alphabets using Data Augmentation.

Research Include.

• Comparing The accuracy Between different Deep Learning Model for ANPR.
• Comparing The efficiency of models and time taken by them for processing the inputs.
• The Trainable Params needed by them and total training time required to run on a machine.
• Computer vision Image Blur issue and how a model can tackle it.
• Confusion between digits like 0 and O also letters like 1 and I.
• Plotting the heatmap of Matrix to see the disturbances and false positives.
• The number of Epochs required for Each model.

Running Models On Local Machine.

Results of Different Models

• Le-Net Model for Different Indian Numberplate Alphabets.

1. Model Accuracy vs Epochs
   
   
2. Model Loss vs Epochs
   
   
3. Model Heatmap for Confusion Matrix
   
   

• MobileNetV2 Model for Different Indian Numberplate Alphabets (Without Data Augmentation).

1. Model Accuracy vs Epochs
   
   
2. Model Loss vs Epochs
   
   
3. Model Heatmap for Confusion Matrix
   
   

• MobileNetV2 Model for Different Indian Numberplate Alphabets (With Data Augmentation).

1. Model Accuracy vs Epochs
   
   
2. Model Loss vs Epochs
   
   
3. Model Heatmap for Confusion Matrix
   
   

• VGG-9 Model for Different Indian Numberplate Alphabets.

1. Model Accuracy vs Epochs
   
   
2. Model Loss vs Epochs
   
   
3. Model Heatmap for Confusion Matrix