In arid and semi-arid regions, pre-emergence soil moisture critically influences cotton seed germination, yet the lack of a crop canopy limits the applicability of conventional remote sensing monitoring. This study investigates Moisture-condensed mulch film as a research subject, exploring it may correlate with subsurface soil moisture conditions. Based on UAV multispectral imagery, a technical pipeline integrating a Spectral-Threshold-Guided Automatic Mask Labeling (STAML) method and a lightweight Dual-Domain Fusion Segmentation Network (DDF-SEGNet) was developed to achieve intelligent recognition of Moisture-condensed mulch film and sub-main pipes. Specifically, STAML leverages spectral threshold characteristics to generate initial masks, significantly reducing the manual annotation burden; DDF-SEGNet adopts a "spatial-dominant, frequency-auxiliary" collaborative architecture, demonstrating favorable applicability in the segmentation task. In experiments conducted in Xinjiang's arid cotton fields, the model achieved favorable segmentation performance with only 0.43 M parameters (Dice = 0.956, mIoU = 0.940) and an inference speed of 168.79 FPS, suggesting its preliminary feasibility for edge-computing deployment. Based on the segmentation results in different irrigation zones from a single-temporal flight, this study suggests Moisture-condensed mulch film's descriptive spatial distribution trends. These findings suggest that, in arid regions, the spatial distribution of Moisture-condensed mulch film may correlate with subsurface soil moisture status, offering a potential exploratory perspective for monitoring soil moisture during critical early growth stages.