This was a group project for PHYS371 at the University of Illinois at Urbana-Champaign. This repository contains the necessary code and documents the results of implementing the YOLOv5 model to identify western corn rootworm beetles on yellow sticky traps. The architecture of YOLOv5 is summarized in the GitHub repository.

Preparing the repository

To train/retrain models and run the notebooks provided with this repository, one needs to create the right environment and install certain dependencies. First cd to the project's top directory and do

export PROJPATH=$PWD

Arduino Code and Drone Attachment

The code and files for the data acquisition from the arduino, the design of the PCB, and the enclosure files, are stored in the Assemblies and Hardware directory. More information about these files is explained in the README file provided in Assemblies and Hardware directory.

Download Data

The dataset used by these studies has been made available to use by Professor Spencer of the University of Illinois at Urbana-Champaign It is available at https://universe.roboflow.com/wcr-beetle-detection/wcr-beetle-traps. Please use the steps provided on the Roboflow website to download the latest version of the dataset. We used the validation part of this dataset as the test images for our paper. Some of these images were incorrectly cropped, and so we discarded them from our results.

Make necessary directories

Run the following commands to make the necessary directories

cd $PROJPATH
mkdir -p crop/data/train/images
mkdir -p crop/data/train/labels
mkdir -p crop/data/val/images
mkdir -p crop/data/val/labels
mkdir -p crop/data/test/images
mkdir -p crop/data/test/labels

Setup necessary environment!

The dependencies for the cropping and simulated data are encompassed in the YOLOv5 model requirements. Run the following commands to set up the environment:

cd $PROJPATH
cd yolov5_model
pip install -r requirements.txt

Data pre-processing

To preprocess the data for the YOLOv5 models, please look at the README file in crop folder.

Training your own models

For each model architecture, we have trained a number of alternate variants and they are hosted in the yolov5_model/runs/train directory. If you are interested in training your own models, please follow the instructions in the README file given in the yolov5_model directory. Please use the beetles.yaml as the .yaml file to train your own model. An example of training the model is:

cd $PROJPATH
cd yolov5_model
python train.py --img 1280 --batch 16 --epochs 100 --data beetles.yaml --weights yolov5x.pt --name NAME --workers 30

The model would be then stored in the 'yolov5_model/runs/train/NAME' folder.

Reproducing image classification results

They rely on avaliability of the pretrained models and the datasets in the way they have been setup in the previous section. You can test unlabeled images by

cd $PROJPATH
cd yolov5_model
python test.py --img 1280 --source path/to/images/ --weights path/to/weights/best.pt --name NAME --workers 30

For full data acquisition, you can use the transfer.py file to output the number of predicted beetles as well as the confidence level into an output.csv file. To run this, run the following commands:

cd $PROJPATH
cd transfer
python transfer.py

Reference

The studies in this repository are compiled and explained in this paper: Building an Arduino Drone Attachment to Detect Western Corn Rootworm Beetles Using YOLOv5

To cite this work, please use the following 'bibtex' entry:

@misc{khot_lin_pasq_pavan_2023,
  title={Building an Arduino Drone Attachment to Detect Western Corn Rootworm Beetles Using YOLOv5},
  author={Khot, Ayush and  Lin, Jeremiah and Pasquesi, Samuel and Pavithran, Pavan},
  year={2023},
  url={https://gitlab.engr.illinois.edu/akhot2/group-01-phys371-sp2023/}
}

Please use the contact information given in https://universe.roboflow.com/wcr-beetle-detection/wcr-beetle-traps to gain permission to use the dataset in research. To cite this dataset, please use the 'bibtex' entry given on the website.

Source Code

Existing resources from publicly available repositories have been adapted to implement the image cropping and YOLOv5 model. We are greatly thankful to the authors of the following works for making these repositories and resources publicly available.

• Part of the image cropping code is obtained from: https://stackoverflow.com/questions/60941012/how-do-i-find-corners-of-a-paper-when-there-are-printed-corners-lines-on-paper-i
• Implementation of YOLOv5 has been largely adopted from: https://github.com/ultralytics/yolov5
• Part of the arduino code has been obtained from the examples for the Adafruit libaries

Contact:

For comments, feedback, and suggestions: Ayush Khot (akhot2@illinois.edu), Jeremiah Lin (jl190@illinois.edu), Samuel Pasquesi (smp6@illinois.edu), and Pavan Pavithran (pavanp2@illinois.edu)