faCRSA: a well-labled dataset for wheat root segmentation

The faCRSA dataset was created to facilitate high-throughput analysis of crop root system architecture (RSA). This dataset is primarily intended for training and evaluating deep learning models for semantic segmentation of roots from soil and is structured in the VOC2007 standard format.The data was collected from an experiment at Nanjing Agricultural University, featuring the wheat cultivar 'Yangmai 16' grown in rhizoboxes. The experiment included control, drought, and waterlogging treatment groups.Image acquisition was performed every two days using a fixed Canon EOS 550D digital camera inside a shooting chamber with integrated LEDs to ensure consistent lighting. The original high-resolution RGB images (1430 × 4500 pixels) were then processed5. Each image was cropped into six smaller patches (715 × 1500 pixels). Subsequently, root and soil pixels were manually annotated and converted to a binary format. To improve model robustness, data augmentation techniques such as random rotation, flipping, and scaling were applied. The final dataset comprises 8,324 images, divided into training (6,997), validation (655), and testing (672) sets.Its application allows for the automated extraction and analysis of RSA traits, enabling research on root plasticity in response to environmental stresses like drought and waterlogging. The complete dataset and associated tools are also publicly available for research use at https://facrsa.aiphenomics.com.