Abstract: To address the challenge of accurately extracting phenotypic parameters from in situ root images collected from minirhizotrons amidst background noise interference, this paper proposes a minirhizotron root phenotypic parameter measurement system based on an improved U-Net model. In the U-Net network, optimized ASPP（Atrous Spatial Pyramid Pooling） and ECA（Efficient Channel Attention） modules are employed to increase the receptive field and enhance the ability to capture detailed features, thereby obtaining precise segmentation images. The experimental results show that the mean intersection over union and mean pixel accuracy of the improved U-Net model are 87. 07% and 91. 85%, which are 2. 49% and 2. 3% higher compared to the original U-Net, respectively. Comparing with measurements obtained using WinRHIZO root analysis software, the determination coefficients for the root length and area are 0. 951 8 and 0. 984 9. respectively. The Spearman correlation coefficients are 0. 972 5 for the root length and 0. 975 7 for root area. This indicates the system′s capability to accurately measure the root length and area.