Precision agriculture relies on efficient crop recognition technology, yet existing semantic segmentation models face dual challenges of real-time performance and accuracy in complex field environments. To address this, this paper proposes a lightweight real-time semantic segmentation network, LCGseg, with the following core innovations: (1) A Context Guided Gate (CGG) mechanism is designed and incorporated into a lightweight pyramid pooling module (CGGPPM) to enhance small-target segmentation through multi-scale feature aggregation; (2) An encoder-decoder architecture is adopted for LCGseg, integrating a Lightweight Convolutional Efficient Multi-scale Attention module (LCEMA) module into skip connections to strengthen the fusion of detail and semantic features. Experiments on the PhenoBench dataset demonstrate that LCGseg outperforms mainstream real-time lightweight models across metrics including IoU, mIoU, and PA while maintaining real-time inference speeds. Ablation studies validate the synergistic effect of the CGGPPM and LCEMA modules, achieving an optimal balance between accuracy and speed. This provides an effective solution for real-time agricultural applications in field settings.