CED-Net: Crops and Weeds Segmentation for Smart Farming Using a Small Cascaded Encoder-Decoder Architecture

Our Contributions

• The paper proposes a new semantic segmentation method called CED-Net (Cascaded Encoder-Decoder Network), specifically designed to differentiate weeds from crops in smart farming applications.
• CED-Net utilizes a small cascaded encoder-decoder architecture, significantly reducing the number of parameters compared to existing methods12.
• The network consists of four independently trained models - two for weed segmentation and two for crop segmentation, allowing for coarse-to-fine predictions12.
• CED-Net outperforms state-of-the-art architectures like U-Net, SegNet, FCN-8s, and DeepLabv3 across various evaluation metrics, including IoU, F1-score, and sensitivity12.
• The proposed network achieves superior results while using only a fraction of the parameters required by other architectures (approximately 1/5.74 of U-Net, 1/5.77 of SegNet, 1/3.04 of FCN-8s, and 1/3.24 of DeepLabv3).

## Paper Available Here