USER STORY: ML Tasks:

[Huddy2022]

As a client I want to predict if a cherry leaf is healthy or contains powdery mildew so that can quickly and accurately identify the presence of powdery mildew in my cherry crop. This will enable me to implement timely interventions and preventive measures, leading to improved disease management, higher yield, and better overall crop health.

[Huddy2022]

Task 1: Count the number of images per label and set.

• Action: Calculate the count of images for each label (healthy and powdery mildew) in the train, test, and validation sets of the cherry leaf dataset.

[Huddy2022]

Task 2: Plot a bar chart for label distribution.

• Action: Create a bar chart visualization to represent the distribution of images across the two labels (healthy and powdery mildew) in the dataset. This will provide a visual comparison of the number of images in each category.

[Huddy2022]

Task 3: Plot a pie chart for dataset distribution.

• Action: Generate a pie chart visualization to illustrate the distribution of images in the cherry leaf dataset across the train, test, and validation sets. This will give an overview of the dataset composition and the proportion of images in each set.

[Huddy2022]

Task 4: Prepare the dataset for machine learning.

• Action: Clean and pre-process the cherry leaf dataset by resizing the images, splitting them into train, test, and validation sets, and applying any necessary data augmentation techniques.

[Huddy2022]

Task 5: Plot augmented images and save class indices.

• Action: Apply image augmentation techniques to the cherry leaf images in the training set, such as rotation, flipping, and zooming, to generate augmented images. Plot a sample of the augmented images to visually inspect the changes made. Additionally, save the class indices mapping for the augmented images, indicating whether each image belongs to the "healthy" or "powdery mildew" class. This information will be useful for training the machine learning model and ensuring that the augmented images are properly labelled.

[Huddy2022]

Task 6: Build a convolutional neural network (CNN) model.

Action: Develop a CNN model using TensorFlow or Keras to perform the classification task of distinguishing between healthy and powdery mildew-contained cherry leaves.

[Huddy2022]

Task 7: Perform hyperparameter optimization.

• Action: Utilize techniques such as grid search or random search to systematically explore various hyperparameter configurations for the CNN model. This includes parameters like learning rate, batch size, number of filters, kernel size, and dropout rate, aiming to maximize accuracy and minimize overfitting or underfitting.

[Huddy2022]

Task 8: Train the CNN model.

• Action: Train the CNN model on the prepared dataset, using the optimized hyperparameters, to learn the patterns and features indicative of healthy and powdery mildew-contained cherry leaves.

[Huddy2022]

Task 9: Plot the learning curve to analyze model performance.

• Action: Visualize and analyze the learning curves to gain insights into the model's performance over epochs. Identify patterns such as convergence, overfitting, or underfitting from the learning curves.

[Huddy2022]

Task 9: Evaluate the trained model.

• Action: Assess the performance of the trained CNN model on the test set by calculating accuracy and loss, then saving the evaluation pickle.

[Huddy2022]

Task 10: Create a confusion matrix.

• Action: Generate a confusion matrix to visualize the model's predictions and identify any misclassifications between healthy and powdery mildew-contained cherry leaves.

[Huddy2022]

Task 10: Create two classification reports.

• Action: Generate two classification reports using the trained model:
  • Classification Report (Model A): Calculate precision, recall, F1-score, and support for each class (healthy and powdery mildew).
  • Classification Report with Macro Avg and Weighted Avg (Model B): Calculate precision, recall, F1-score, and support for each class, along with macro-averaged and weighted-averaged metrics.

[Huddy2022]

Task 11: Predict class probabilities for a random image.

• Action:

  • Select a random image from the test set.
  • Convert the image to an array and preprocess it for prediction.
  • Use the trained model to predict the class probabilities of the image.
  • Display the predicted class probabilities along with the corresponding labels (healthy or powdery mildew).
  • Assess the model's performance by comparing the predicted probabilities with the true label of the image.
  • Analyze the predicted probabilities to evaluate the model's ability to differentiate between healthy cherry leaves and those with powdery mildew.