基于深度学习的遥感影像地物分类方法、系统、设备及介质
By preprocessing based on quantile normalization and overlapping cropping, combined with a sliding window attention mechanism and a dual-path supervised segmentation model, the network parameters are optimized, solving the problem of efficient classification of large-scale remote sensing images and achieving high-precision and robust end-to-end automated processing.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INNER MONGOLIA UNIVERSITY
- Filing Date
- 2026-02-06
- Publication Date
- 2026-07-17
AI Technical Summary
Existing deep learning methods are computationally expensive and memory-intensive when processing large-scale, high-resolution remote sensing images. They also suffer from insufficient classification accuracy and robustness, low automation, and difficulty in adapting to the rapid processing needs of images from different regions and sensors.
By employing statistical normalization based on preset quantiles and overlapping grid clipping, combined with a sliding window attention mechanism and a hierarchical Transformer encoder, a dual-path supervised segmentation model is introduced. The network parameters are optimized through a composite loss function and learning rate strategy to achieve efficient land cover classification.
It enables efficient and lossless processing of large-size remote sensing images, improves classification accuracy and robustness, ensures end-to-end automation from data preprocessing to inference, and adapts to the rapid processing of images from different regions and sensors.
Smart Images
Figure CN122067014B_ABST