Cotton farmers across India face significant crop losses due to bollworm infestations. To manage the problem, farmers have been encouraged to install pheromone traps that capture male bollworms, which are capable of reproduction. By counting the number of bollworms in these traps, farmers can estimate potential infestations and make informed decisions about pesticide use to minimize future damage.
Wadhwani AI developed a mobile app to assist farmers, allowing them to photograph trap catches and receive recommendations based on machine-generated counts of pests. The app’s object detection model identifies two types of bollworms, but the accuracy of counting remained an area for improvement. To address this, Wadhwani AI, in collaboration with Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) GmbH, hosted a competition to build a more accurate machine learning solution for counting bollworms in images.
Out of 705 data scientists worldwide, I (platform name: flamethrower) working with Team FlameTurbo developed the 3rd place winning solution. The project involved designing an object detection model capable of reliably counting pests in varied conditions, ensuring practical utility for farmers in the field. This challenge highlighted the power of applied machine learning to support agriculture, improve decision-making, and reduce crop losses.
Project Link:
Leave a Reply