Image segmentation of petrographic thin sections is a critical task in oil and gas exploration and geological research. This paper proposes a lightweight Cascaded Multi-Receptive Fields U-Net (CMRF-UNet) model to improve segmentation accuracy in petrographic thin-section images, reducing both parameter count and computational complexity. The method incorporates a cascaded multi-receptive field module based on the UNet architecture. By efficiently fusing depth-separable convolution with redundant feature channels, it reduces the model's parameter count while enhancing its multi-scale feature representation. Additionally, a hybrid loss function combining Dice loss and cross-entropy loss is constructed to mitigate the impact of class imbalance on the segmentation of small targets. To evaluate the proposed method, experiments were conducted on the petrographic thin-section image dataset using CMRF-UNet. Experimental results show that CMRF-UNet outperforms conventional models in segmentation accuracy while reducing the model's parameter count by 92.79%. The proposed CMRF-UNet method provides an effective solution for the image segmentation of petrographic thin sections and holds significant engineering application value.