Leaf-Septoria-Detection-On-tomato-leaves

Leaf Septoria is a disease that effects tomato plants quite often.

We used CNN here to detect that disease.

More on the project

Augmentation methods used:

1. Rotation by 30 degrees
2. Shifting along the width
3. Shifting along the height
4. Shearing the image
5. Horizontal & Vertical flip
6. Introducing different lighting conditions

Fill method used: 'nearest'

Model Description:

We have used a Convolutional Neural Network with 3 Convolution and 3 MaxPool layers.

Dropout was introduced in the Dense layers as follows:

1. 40% Dropout before Flattening
2. 15% Droput before the first hidden layer
3. 25% Dropout before the Output layer

Information on the dataset:

Actual Dataset: https://www.kaggle.com/emmarex/plantdisease

Download the dataset, from there extract these two files: Tomato_Septoria_leaf_spot, Tomato_healthy

Now, we create train & validation sets manually.

There are 3362 images in total.

So, if you want valid_set size = 0.2 i.e. 20% of the data, then you'll have to select a total of 672 images for the validation set.

Steps:

1. Make a folder valid and 2 other folders Healthy leaf and Septoria Leaf.
2. Then we randomly select 312 images from Tomato_healthy folder and move them into Healthy leaf folder.
3. We randomly select 360 images from Tomato_Septoria_leaf_spot folder and move them into Septoria Leaf folder.
4. Then move the Septoria Leaf & Healthy leaf into the valid folder.

Your validation dataset is complete!

Now, make 2 more folders of the same name: Septoria Leaf & Healthy Leaf

Move the remaining images into the respective folders and place these two folders in a new folder train

Your dataset is complete!

Upload the train and valid data into your Google Drive and mention the paths after the drive is mounted on Google Colab.