Grain-Detection

Model: Grain Classification Model

Description:

This model is design to classify 4 different types of grains: noodles, rice, couscous, and oatmeal by using the FastAI library with Fastbook and ResNet-18 convolutional neural network architecture. <br> The model files can be found in the "Files and Versions" section.

Training Data

• Limited dataset of 400 labeled images of grains, with 100 images for each category.
• Obtained using DuckDuckGo Images API
• Resized to 128x128 pixels to reduce storage and computation usage.
• 3 additional training epochs are performed to fine-tune the model for grain classification task.

Metrics

Model performance are evaluated using confusion matrix. Confusion matrix measures the metric of accuracy and precision. The prediction is in 4 categories:

• True positive: Model correctly predicted positive class.
• False Positive: Model incorrectly predicted when actual is negative class.
• True Negative: Model correctly predict negative class.
• False Negative: Model incorrectly predict negative class when actual is positive.

Results

In the images below, you can see that there are different shades.

• The darker shade represents correct predictions
• The lighter shade represents incorrect predictions.

<table> <tr> <td>First training:</td> <td>Second training:</td> </tr> <tr> <td><img src="1GrainFirstTraining.png" alt="First training" width="300"/></td> <td><img src="2GrainRetrain.png" alt="Second training" width="300"/></td> </tr> </table>

We can see that the model has improved after cleaning the data.

• it improved in predicting the images correctly.
• there is better accuracy in the second training compared to the first training. The number of darker shades is higher.
• there is less confusion in the second training compared to the first training. The number of lighter shades is lower.