Integrating CNN deep learning and biomass correlation approaches for paddy land detection and yield estimation in north central province, Sri Lanka

Abstract

Accurate paddy land monitoring and production forecasting are critical for ensuring food security and making informed import decisions. This research focuses on integration of Convolutional Neural Network (CNN) deep learning models with vegetation index-based analyses to enhance paddy land identification and yield estimation in the North Central Province of Sri Lanka. The research design outlined involves two overall stages. First stage involves two steps. First, CNN models are applied to high-resolution satellite imagery to detect paddy lands in the region accurately. Second, the Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI) are utilized to classify active and inactive paddy fields during cultivation seasons. The second stage is the production estimation using a biomass correlation approach by employing NDVI and the Enhanced Vegetation Index (EVI) to predicting yield before harvesting. The study addresses a major problem in Sri Lanka’s agricultural sector—inaccuracy of data, which typically leads to inefficient decisions in rice import and compromises food security at national level. By providing timely and accurate information on active paddy lands and expected production, the proposed method offers a scientific basis for better planning and policy-making in the agricultural sector.