Intelligent recognition and vectorization automatic extraction technology of mining area buildings based on high-resolution remote sensing images

This study developed an automatic identification and vectorization extraction technology of buildings based on high-resolution remote sensing images, combining image enhancement, object classification and transfer learning optimization to achieve precise extraction of buildings. Through multi-scale image fusion using principal component analysis (PCA) and wavelet transform, significant improvement in detail resolution was achieved, making building boundaries clearly visible for visual input support to the classification model. The classification models based on Support Vector Machine (SVM) and Random Forest performed well in building recognition and change detection, especially with stronger robustness and accuracy demonstrated by the random forest model. With transfer learning, the ability of extracting buildings from complex environments significantly improved, resulting in higher precision and overall accuracy reaching 83.12%. It is shown that application of transfer learning reduces reliance on large amounts of annotated data in high-resolution imaging processing. Future work can further combine deep learning models such as U-Net and DenseNet along with more sophisticated multi-scale image fusion techniques to enhance the accuracy and broad applicability of building extraction. This research provides a viable solution for automated building detection in remote sensing imagery.