PhytNet Training Framework

This repository contains the implementation for training the PhytNetV0 model, a convolutional neural network designed specifically for classifying plant diseases from images. The training script supports both standard training and hyperparameter optimization through Weights & Biases (WandB) sweeps. The predefined optimisation sweep should be run to customise the model architecture for your dataset before final training. While boilerplate dataloading code etc. has been provided, the train() function has been left blank for you to insert your own code. The original published PhytNet paper can be found here: https://bsapubs.onlinelibrary.wiley.com/doi/full/10.1002/aps3.11620

Abstract

Abstract Premise Automated disease, weed, and crop classification with computer vision will be invaluable in the future of agriculture. However, existing model architectures like ResNet, EfficientNet, and ConvNeXt often underperform on smaller, specialised datasets typical of such projects.

Methods We address this gap with informed data collection and the development of a new convolutional neural network architecture, PhytNet. Utilising a novel dataset of infrared cocoa tree images, we demonstrate PhytNet's development and compare its performance with existing architectures. Data collection was informed by spectroscopy data, which provided useful insights into the spectral characteristics of cocoa trees. Cocoa was chosen as a focal species due to the diverse pathology of its diseases, which pose significant challenges for detection.

Results ResNet18 showed some signs of overfitting, while EfficientNet variants showed distinct signs of overfitting. By contrast, PhytNet displayed excellent attention to relevant features, almost no overfitting, and an exceptionally low computation cost of 1.19 GFLOPS.

Conclusions We show that PhytNet is a promising candidate for rapid disease or plant classification and for precise localisation of disease symptoms for autonomous systems. We also show that the most informative light spectra for detecting cocoa disease are outside the visible spectrum and that efforts to detect disease in cocoa should be focused on local symptoms, rather than the systemic effects of disease.

Prerequisites

Before you begin, ensure you have met the following requirements:

Python 3.x

PyTorch

torchvision

Weights & Biases (wandb)

PyYAML

You can install the necessary libraries using pip:

pip install torch torchvision wandb pyyaml

Configuration

Hyperparameters and training configurations can be adjusted in two ways:

Directly in the script by modifying the config dictionary.

Through a YAML file for WandB sweeps (see PhytNet-Sweep-config.yml for an example).

Dataset

Your dataset should be organized into separate train and val directories, each containing subdirectories for each class. Place your dataset in a known directory and update the data_dir variable in the script accordingly.

Usage

To run the training script, simply execute:

python train_phytnet.py

To initiate a WandB sweep, first ensure you have a PhytNet-Sweep-config.yml file configured with your desired sweep parameters. Then, set sweep_config in the script to the path of your YAML file.

Hyperparameter Tuning

When using WandB for hyperparameter tuning, make sure you're logged into WandB and have configured PhytNet-Sweep-config.yml with your desired sweep parameters. The script will automatically handle the sweep initialization and execution.

Contributing

Contributions to the PhytNet architecture are welcome. Before contributing, please review the contribution guidelines.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

This project is built using PyTorch and Weights & Biases for model training and hyperparameter optimization.