Fertilizer App: Potato Disease Identification 🌱🦠

The Fertilizer App is designed to help identify common potato leaf diseases using deep learning models. Currently, the model is trained to classify potato leaves into three categories:

• Early Blight 🌿
• Late Blight 🍂
• Healthy ✅

In the future, this app will be generalized to support more crop diseases and broader agricultural applications.

Features 🎉

• Disease Detection: Uses a TensorFlow model to predict diseases in potato leaves from uploaded images.
• Model Inference: Upload an image of a potato leaf, and the model will predict if the leaf is healthy or affected by diseases like Early Blight or Late Blight.
• Future Expansion: The model is being generalized to support other crops and their respective diseases.

CNN Architecture🍃

The deep learning model used is based on a Convolutional Neural Network (CNN), which is highly effective for image classification tasks.

Key Components

1. Preprocessing Layers:

   • Resizing: The images are resized to 256x256 pixels to standardize the input size for the network.
   • Rescaling: Image pixel values are rescaled by dividing by 255, converting the pixel values into the range [0, 1].

2. Data Augmentation:

   • Random Flip: Images are randomly flipped both horizontally and vertically to simulate various real-world conditions and improve model generalization.
   • Random Rotation: The images are randomly rotated by up to 20% to enhance the model’s ability to recognize potato diseases under different orientations.

3. Convolutional Layers:

   • Conv2D: Several convolutional layers with 32 and 64 filters are used to extract important features from the potato leaf images.
   • ReLU Activation: The ReLU activation function is applied after each convolutional layer to introduce non-linearity and help the model learn complex patterns in the images.

4. Max-Pooling Layers:

   • MaxPooling2D: Max-pooling layers are used after each convolutional layer with a pool size of 2x2 to reduce the spatial dimensions of the feature maps, retaining only the most relevant information.

5. Fully Connected Layers:

   • Flatten: After the convolutional and pooling layers, the feature maps are flattened into a 1D vector to be fed into fully connected layers.
   • Dense Layer (64 units): A fully connected layer with 64 units is used to connect the features extracted by the convolutional layers to the final output.
   • Output Layer: The output layer has 3 units, one for each class (Early Blight, Late Blight, Healthy), with a softmax activation to output the class probabilities.

Summary of Layers:

• Input: Resized 256x256x3 (RGB) image
• Conv2D Layers: 32 and 64 filters with ReLU activation
• MaxPooling2D Layers: Pooling with a 2x2 filter size
• Dense Layers: 64 units followed by the final softmax output layer

This architecture was implemented using TensorFlow and Keras to ensure efficient model training and inference. The model was trained on potato leaf images and uses data augmentation techniques to improve generalization and robustness.