Figure 14 | Scientific Reports

Figure 14

From: Generating multi-temporal landslide inventories through a general deep transfer learning strategy using HR EO data

Figure 14

Schematic diagram of the transfer learning approach undertaken in this study. Weights are obtained from the first study area and then used along with training samples in the second study area for assistance while training in the second area. This is followed until the last study area where the model is trained using weights from the first three areas and samples of the fourth, respectively. An additional advantage of this modelling layout is that new study areas can be added at any moment producing a possible further model improvement in accuracy, generalization, and overfitting.

Back to article page