Soil whitening in cold-region rice seedbeds is visually associated with surface drying and moisture variation. The timely and objective monitoring of soil surface conditions is therefore important for seedbed management. In response to the inefficiencies of manual scouting and the limitations of conventional threshold-based methods under varying illumination and complex soil textures, this study presents a seedbed soil whitening analysis method that combines an enhanced DAC-UNet for semantic segmentation with colour feature analysis. First, a binary segmentation dataset of soil and background was created using RGB seedbed images. Within the U-Net framework, deformable convolution, ASPP++ multi-scale feature aggregation, and the CBAM attention mechanism were introduced to improve the model’s representation of irregular boundaries, scale variations, and complex illumination conditions. Comparative experiments demonstrated that the proposed model achieves 90.63% MIoU, 94.82% mPA, and 97.52% accuracy on the soil segmentation task. Based on the segmented soil region, a Whiteness Index (WI) was formulated to characterize soil surface whitening and generate whitening heatmaps. This enables quantitative description and spatial visualization of whitening characteristics within the soil region. Experimental results showed that the proposed method can effectively capture visual differences among different soil whitening states and provide intuitive visual reference information for soil surface condition analysis in cold-region rice seedbeds.