Efficient Root Segmentation Using Multi-ResUnet Architecture

Root segmentation in plant images plays a crucial role in agricultural research and automated farming systems. This work introduces Multi-ResUNet, an efficient architecture for root segmentation leveraging multi-resolution features and a modified UNet framework. The Multi-ResUNet model enhances segmentation accuracy by integrating skip connections across multiple scales, facilitating the capture of both local and global contextual information crucial for accurate root delineation. The proposed model obtains improved performance compared to existing models with an IoU of 0.67 incase of chicory root dataset.