The recognition and extraction of land features in remote sensing images based on AI big models has become a key technology to deepen its application in the field of energy and power

and model training relies on massive sample data.This article addresses the issues of high annotation workload

low automation level

and low quality when establishing a sample library.By studying the factors that affect sample production and the mapping form in inference results

a sample library architecture is proposed that is targeted according to the target category.A SAM model based on point and line representation is used to achieve high-precision semi-automatic sample rapid annotation

and the construction effect of the sample library is verified by combining a small sample learning model.The results indicate that using the constructed sample library for model training

the inference edges of various elements are accurate and have comparable accuracy

indicating that the sample library distribution is reasonable

the structure is good

and the annotation accuracy is high.This method can improve the intelligent production of samples

reduce manual workload

and ensure the quality of sample libraries

providing reference ideas for the establishment of remote sensing image AI model sample libraries.