The detection of strip steel surface defects is critical to ensure the quality of strip steel productions. At present, many deep-learning-based methods have been presented and achieve outstanding performance. However, most of these methods ignore the frequency information among defect areas, which plays an important role in defect detection. This paper proposes a deep learning method to further improve defects segmentation effect based on existing methods, called Low-Pass U-Net. Since most defects on strip steel are located in high frequency areas, we implement a low-pass filter before downsampling in encoder, which prevents aliasing and separates out high frequency information. The high frequency feature is transferred into decoder to assist segmentation. Following previous works, we propose adaptive variance gaussian low-pass layer to generate different filters according to each spatial location of feature map with fewer computing resource. Besides, to detect defects at significantly different scales, an improved Hypercolumn module is adopted at the end of decoder to upsample and fuse the feature maps in different resolutions, where Subpixel replaces the bilinear interpolation to refine the upsampled results. The proposed method is validated on practical datasets and achieves considerable performance improvement (with best Dice Coefficient 0.903), which demonstrates the effectiveness of Low-Pass U-Net. The introduction of AVGLFL layer results in 3% increase of Dice Coefficient in a comparative inference time, which achieves a balance in performance, inference time and complexity.