Abstract:
Cotton cultivation is pivotal to Pakistan’s economy, contributing significantly to its GDP and
accounting for a substantial portion of its export earnings. However, the country currently faces
a steep decline in cotton production due to challenges such as reduced sowing area, adverse
weather conditions, and pest attacks. This underscores the urgent need for enhanced agricul tural data to support effective decision-making and improve cotton yield forecasts, crop health
monitoring, and policy formulation.
This study presents a comprehensive approach to cotton crop mapping in Pakistan using ad vanced remote sensing techniques and machine learning, specifically through the application of
semantic segmentation. It capitalizes on multispectral imagery collected via Unmanned Aerial
Vehicle (UAV) sensors and Sentinel-2 satellite data to map cotton fields with high precision. By
correlating data from these two sources, the research confirms the reliability of satellite imagery
in agricultural applications. The methodology employs a transfer learning strategy that begins
with training a model using a large dataset of Uzbekistan cotton. This model is then fine-tuned
on labeled data from Pakistan for the year 2023 to adapt to local agricultural conditions. To ad dress the challenge of sparse data for the previous year, active learning is utilized to efficiently
create a labeled dataset for the 2022 cotton crop in Pakistan.
The study demonstrates a method that leverages data from data-rich regions to enhance pre dictions in data-scarce regions like Pakistan. This approach not only improves the accuracy of
crop mapping but also reduces the dependency on extensive manual data labeling, thus provid ing an efficient solution to enhance the agricultural data infrastructure in developing countries.
Through this research, we advocate for the combined use of remote sensing, machine learning,
transfer learning, and active learning to address the critical challenges faced by the agricultural
sector in Pakistan, potentially leading to more informed decisions and sustainable agricultural
practices.
