SEDLNet: An unsupervised precise lightweight extraction method for farmland areas.

•Proposed a neural network model for unsupervised segmentation of farmland areas.•In this model, channel attention mechanism and double loss function were introduced to improve feature extraction and reduce network depth.•The mIOU of this model reached 0.7675 in terms of farmland extraction. In the tree-belt farmland environment, the mIOU of this method reached 0.7475.•Line labels can be introduced as false labels to improve the extraction accuracy.