Remote Sensing Scene Classification via Spatial Relationship Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional remote sensing image analysis methods fail to effectively analyze spatial relationships among images, leading to inferior performance in classifying urban scenes compared to natural images.
Innovation Solution
A method that cuts remote sensing images into sub-images, performs visual information coding, and uses a crossing transfer unit to extract and fuse regional and long-range spatial relationships, followed by dimensionality reduction and softmax classification to improve scene classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If deep learning models based on natural image analysis are used for remote sensing image classification, then the model can process images efficiently, but the spatial relationships specific to remote sensing images cannot be modeled effectively
Solution Approach 1:
The patent segments the remote sensing image into multiple sub-images, processes each sub-image independently through visual information coding to extract features, and then reconstructs the spatial relationships between sub-images. This segmentation approach enables efficient processing of large remote sensing images while preserving spatial relationship information that would be lost in holistic processing.
Solution Approach 2:
The patent introduces spatial relationship encoding as an intermediary mechanism between the deep learning model and the remote sensing image data. By encoding spatial relationships separately and integrating them with feature extraction results, the model can leverage both the efficiency of deep learning and the spatial specificity of remote sensing images without requiring complete model redesign.
2Ease of manufacture
If relationships between images are discarded when building a data set for model training, then the data preparation process is simplified, but the performance of the deep learning model on remote sensing images deteriorates
Solution Approach 1:
The patent performs preliminary encoding of spatial relationships during the data preparation phase, storing spatial relationship information alongside image features. This preliminary action maintains data preparation simplicity while ensuring that spatial relationship information is preserved and available for model training, thereby improving model performance without significantly complicating the data preparation process.
3Measurement precision
If remote sensing images are processed at full resolution, then classification accuracy is improved, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent divides large remote sensing images into smaller sub-images for independent processing. This segmentation reduces the computational complexity and memory requirements for each processing unit while maintaining overall classification accuracy through proper integration of sub-image results and spatial relationship encoding.
Solution Approach 2:
The patent applies visual information coding selectively to extract essential features from sub-images rather than processing all image data at full resolution. This partial action approach focuses computational resources on the most discriminative features while maintaining classification accuracy, avoiding the need to process every pixel at full resolution.
Data Source
AI summary
An urban remote sensing image scene classification method in consideration of spatial relationships is provided and includes following steps of: cutting a remote sensing image into sub-images in an even and non-overlapping manner; performing a visual information coding on each of the sub-images to obtain a feature image Fv; inputting the feature image Fv into a crossing transfer unit to obtain hierarchical spatial characteristics; performing convolution of dimensionality reduction on the hierarchical spatial characteristics to obtain dimensionality-reduced hierarchical spatial characteristics; and performing a softmax model based classification on the dimensionality-reduced hierarchical spatial characteristics to obtain a classification result. The method comprehensively considers the role of two kinds of spatial relationships being regional spatial relationship and long-range spatial relationship in classification, and designs three paths in a crossing transfer unit for relationships fusion, thereby obtaining a better urban remote sensing image scene classification result.