An Indian annotated weed dataset for computer vision tasks in precision farming

Area of Science:

• Agricultural Science
• Computer Vision
• Data Science

Background:

• Weed infestations pose a significant threat to Indian agriculture, causing an estimated 45% annual productivity loss.
• Traditional weeding methods are labor-intensive and costly for smallholder farmers.
• Increasing herbicide resistance in weeds necessitates advanced management solutions.

Purpose of the Study:

• To introduce the MH-Weed16 dataset for advancing artificial intelligence and computer vision in agriculture.
• To provide a comprehensive resource for developing automated weed detection and management systems.
• To support precision farming initiatives and enhance sustainable agricultural practices.

Main Methods:

• A dataset of 25,972 images, MH-Weed16, was created from real agricultural fields in Maharashtra.
• The dataset features 16 different weed species, annotated by agricultural experts.
• A subset of 7,577 images includes both crops and weeds, captured from a top-down perspective for accurate area estimation.

Main Results:

• The MH-Weed16 dataset comprises 25,972 images with 16 annotated weed species.
• Specific focus on 7,577 images showing crop-weed interactions from a top view.
• The dataset is poised to be a valuable resource for computer vision applications in precision agriculture.

Conclusions:

• The MH-Weed16 dataset is a crucial step towards integrating AI and computer vision for effective weed management.
• This resource will facilitate the development of technologies that reduce crop loss and improve farming efficiency.
• The dataset supports the transition to more sustainable and technologically advanced agricultural practices.