About

---

This is a deep learning model built using CNN.

An image classifier used to classify carrots and apples (a group that possesses a stark difference hence yielding highly accurate results).

This is also my first attempt on working with deep learning models.

Be kind :)

Libraries and components used

---

1. tensorflow
2. os
3. cv2
4. imghdr
5. matplotlib
6. numpy
7. tensorflow.keras.models - Sequential
8. tensorflow.keras.layers - Conv2D
9. tensorflow.keras.layers - MaxPooling2D
10. tensorflow.keras.layers - Dense
11. tensorflow.keras.layers - Flatten

Important!

---

• Since the system I use couldnt run the notebook in a short span of time, Google Colab was used to run the notebook.

• Change all the directories accordingly for proper functioning.

• Datasets.zip contains two folders, Test Data and data. Test Data contains test data (4 files), data contains the training data. Obviously change them if you want to tweak the model to do something else

Clarifications and other necessary information

---

• 70-20-10 rule is used here, 70% of the dataset for training, 20% for validation, 10% for testing

1. Convolution Layers

• The first Conv2D layer has 16 filters of size 3x3, with a ReLU activation function. It's the input layer, taking input images of size 256x256 with 3 channels (RGB)

• The second Conv2D layer has 32 filters of size 3x3 with ReLU activation

• The third Conv2D layer has 16 filters of size 3x3 with ReLU activation.

• Each convolutional layer is followed by a MaxPooling2D layer which reduces the spatial dimensions of the feature maps by taking the maximum value within a specified window (typically 2x2)

2. Flatten layer

• After the convolutional layers, a Flatten layer is added to convert the 3D feature maps into a 1D feature vector, which can be fed into a fully connected layer.

3. Dense Layers

• There is one Dense (fully connected) layer with 256 neurons and ReLU activation.

• The final output layer is also a Dense layer with a single neuron and a sigmoid activation function. This is common for binary classification tasks where the output should be between 0 and 1.

4. Compilation

• The model is compiled using the Adam optimizer and binary cross-entropy loss function, which is suitable for binary classification tasks.

• The metric being tracked during training is accuracy.