Research on the application of remote sensing technology combined with deep learning in the recognition of landscape features of Jizu Mountain

The application of deep learning in remote sensing has opened new possibilities for automatic landscape interpretation, particularly in complex mountainous regions where natural and cultural features coexist. This paper presents a semantic segmentation approach that integrates multi-source satellite imagery with an enhanced ResNet architecture tailored for high-resolution landscape classification. The model introduces spatial attention and multi-scale feature fusion to improve the identification of structures such as temples, vegetation-covered slopes, and sky regions with dense cloud cover. Using data from the Jizu Mountain area in Yunnan, China, the method was evaluated against both traditional machine learning algorithms and standard convolutional models. Results indicate consistent improvements in classification accuracy, precision, recall, and intersection-over-union (IoU), especially in areas with irregular terrain and overlapping land covers. The proposed method demonstrates good generalizability and practical value for use in cultural heritage monitoring and geospatial analysis across similarly complex environments.