PlantDiseaseDetection

Overview

Creating a fast and cost effective way to identify plant diseases using Convolutional Neural Network. Two approaches are applied to address this problem. The first approach uses 5 layers convolutional neural network with Keras. The second approach applies ResNet 152 Deep Neural Network with PyTorch. 90% accuracy obtained

Dataset

Fig. 1 Kaggle PlantVillage dataset

Environment

Python version: 3.X

Develop platform: Google Colab

Prerequisite

The libraries used include: nltk, sklearn, numpy, matplotlib.pyplot, pandas, spacy, seaborn. Dataset: PlantVillage Dataset

How to run

Both Keras.ipynb and Pytorch.ipynb were developed on Google Colab, so it is easy to open it through Colab and run the code cell sequently.

Keras: Open the code file Keras --- main.ipynb Testing Runtime --- Run all Training (Optional): Change TRAIN to True. --- Runtime --- Run all

Pytorch: Open the code file Pytorch--- main.ipynb Testing Runtime --- Run all (Here you may encounter a couple of bugs

1. Need to switch to GPU.

2, Import Error.

    Cannot import name ‘isStringType’’

Training (Optional): Change TRAIN to True. --- Runtime --- Run all --- Enter the authorization code with your UOttawa account

Instructions to fix the previous bugs

1. Need to switch to GPU. Edit --- Notebook settings --- Hardware accelerator --- Select ‘GPU’ --- Save --- Runtime --- Run all

2. Import Error. Cannot import name ‘isStringType’ Runtime --- Restart runtime --- Select Cell 2 --- Run after

Results

Fig. 2 Training and testing accuracy for both approaches