A Multi-Varietal, Multi-stage Rice Phenotyping Dataset (RicePheno-3V4S)

The RicePheno-3V4S dataset has been collected from experimental and cultivated paddy fields at the SRM College of Agricultural Sciences, Minnal Chiattamur during the RABI season under controlled agronomic practices. The research fields were cultivated following the recommended agronomic guidelines for each variety by the experts. The experimental area was comprised 30 cents of land in total. To improve dataset diversity and model generalization, RicePheno-3V4S includes multiple rice varieties such as CO51, CR1009, and IWPonni. 10 cents of land area were used for each variety. All images were manually annotated and agronomically verified to ensure accurate stage labelling and biological consistency Additionally, full phenological tracking was achieved using images obtained across four key stages like Active Tillering, Panicle Initiation, Flowering, and Harvesting crucial for yield determination. Our dataset allows synchronization of multiple sensors by simultaneously capturing images with a UAV, DSLR and mobile cameras on the same site providing different views of the same landscape. The multi-variety, multi-stage and multi-sensor design of RicePheno-3V4S makes it a robust benchmark dataset for agricultural artificial intelligence, precision farming, crop monitoring and phenology-based yield prediction systems. The dataset with 512 x 512 image resolution is particularly suitable for developing generalized and deployable deep learning models for real-world smart agriculture applications.