A comprehensive dataset of rice field weed detection from Bangladesh

Abstract

In agricultural research, particularly concerning rice cultivation, the presence of weeds within rice fields is acknowledged as a significant contributor to both diminished crop quality and increased production costs. Rice fields, due to their inherently moist environment, offer ideal conditions for weed proliferation. Traditionally, the control of these weeds has been managed through labor-intensive manual methods. However, as the agricultural sector evolves, there is a notable pivot towards leveraging advanced technological solutions, including deep learning and machine learning. The efficacy of these technologies hinges on the availability of high-quality, relevant data. To address this, a comprehensive dataset comprising 3632 high-resolution RGB images has been developed. This dataset is designed to capture a diverse range of weed species, specifically 11 types that are frequently found in rice fields. The diversity of the dataset ensures that machine learning models trained using this data can effectively identify and differentiate between desired and undesired plant species. While the dataset predominantly includes images from Bangladesh, the weed species it documents are commonly found across various global rice-growing regions, enhancing the dataset's applicability in different agricultural settings.