PadYi: Ensembled CNN-Transformer Attention Based Framework for Optimizing Paddy Cultivation Management and Yield Estimation

Paddy Rice cultivation in paddy fields, encompassing approximately 167 million hectares globally, serves as a critical staple for over half of the world’s population and plays a pivotal role in global food security. However, this agricultural practice faces significant challenges, including water scarcity, climate change impacts, and inefficient farming methods, which collectively jeopardize its sustainability and productivity. This research introduces a novel self-curated dataset, PadiFier, and presents a comprehensive multi-stage framework that leverages advanced machine learning techniques, including Convolutional Neural Networks and Transformers. The framework incorporates state-of-the-art architectures such as ResNet18, EfficientNet, SEResNeXt, DeiT, and MobileNetV3 to enhance paddy cultivation management. Our multi-stage model exhibits state-of-the-art performance, achieving high aggregate accuracy across all stages, comparable to cutting-edge technologies in agricultural image analysis. By providing real-time, precise information on crop status, health, and potential yield, this innovative technology enables farmers to make data-driven decisions, optimize resource allocation, and implement precision agriculture techniques effectively.