Track Description
- Akshat Nitin Savla
In this hackathon track, participants will tackle a fundamental challenge in precision agriculture: distinguishing crops from weeds. Distinguishing crops from weeds is a core target for many companies who make robots or precision sprayers for weed control, making it a crucial area of innovation in agricultural technology.
Objective
Participants will label crops and weeds in provided images, then train a computer vision model to automatically detect and classify these plants. This challenge is designed to be accessible to all participants - no coding is required, and the only tool needed is a web browser.
Task
Annotate the crops and weeds in the provided dataset.
Use the labeled data to train a machine learning model.
Test and refine your model to optimize performance.
You will be provided with a dataset consisting of images containing crops (wheat) and weeds (ryegrass).
Useful link: https://www.superannotate.com/blog/introduction-to-image-annotation
Tools
All work will be done using Roboflow, a platform for computer vision tasks. Roboflow integrates all the processes involved in a computer vision project in one place, from data annotation to deploying a final trained model.
A guide to using Roboflow can be found here: https://precision-sustainable-ag.atlassian.net/wiki/x/A4CYKg The guide also describes steps to set up a free account which grants more training credits.
Dataset
Participants will work with a dataset of 400 images featuring two primary plant species: winter wheat (crops) and Italian ryegrass (weeds). Each image contains varying densities of both crops and weeds, simulating real-world agricultural scenarios.
The objective is to accurately label these species within each image, creating a clear distinction between the crop and weed classes. Accurate annotation is critical for the performance of the model in distinguishing between these two plant species.
The dataset can be downloaded from the following link:
Sample images from the dataset
Evaluation
The final evaluation of a participating team will be based on multiple factors, ranging from model performance to the techniques used during the hackathon. A detailed rubric for the evaluation and process of submission can be found here: https://precision-sustainable-ag.atlassian.net/wiki/x/CIC9Kg