Detecting Weeds and Crops Using Computer Vision on Drone Imagery

Background

Persistent herbicides are known to contaminate the environment, harm non-target organisms, and pose risks to human health. Addressing these challenges is critical for promoting sustainable agricultural practices. Skymaps, an agtech startup leveraging remote sensing technologies and advanced image analysis, partnered with Omdena to tackle this pressing issue. Their shared mission focused on improving agricultural interventions to achieve sustainable agricultural production.

Objective

The primary goal of this project was to develop a computer vision solution capable of detecting and identifying different weed species and crop types. By doing so, the solution aims to significantly reduce the reliance on herbicides, thereby supporting environmentally friendly farming practices.

Approach

To achieve the project objectives, the team adopted a multifaceted approach:

• Data Collection: Annotated weed and crop samples from the field to enhance the model’s performance.
• Model Development: Developed a state-of-the-art computer vision model tailored to integrate seamlessly with the Skymaps application.
• Advanced Features: Implemented user-selectable options for crops to refine the model’s predictions and focus on the most probable weed species.

Results and Impact

The project delivered a cutting-edge ML model with the following capabilities:

• Detection and identification of various weed species and crops, including corn, cereal, and sunflower.
• Integration with Skymaps’ application, enabling real-time analysis and user interaction.
• Enhanced annotation database to improve the model’s accuracy and reliability.

Impact:

• Reduced reliance on herbicides, leading to environmentally friendly farming practices.
• Minimized contamination of the environment and harm to non-target organisms.
• Support for sustainable agricultural production.

Future Implications

The success of this project sets the stage for future advancements in precision farming. The developed model can be further refined with additional data and adapted to other agricultural contexts. These findings could also inform policy changes aimed at promoting sustainable farming practices and reducing chemical interventions.

By leveraging technology, this initiative represents a significant step toward a more sustainable and efficient agricultural future.