Random Block Semantic Segmentation for High-Resolution Remote Sensing Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing semantic segmentation methods for high-resolution remote sensing images face GPU memory overflow issues and struggle to accurately identify objects with similar colors due to their large data size and multispectral nature.
Innovation Solution
The method partitions high-resolution remote sensing images into random blocks, extracts mono-spectral feature data from each block, performs semantic segmentation using a supervised network that incorporates neighboring block data, and fuses the results with trained weights to generate a fused semantic segmentation probability plot.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If high-resolution remote sensing images are directly input into existing semantic segmentation networks, then the network can process the complete image data, but GPU memory overflow occurs due to the large data size
Solution Approach 1:
The patent divides the high-resolution remote sensing image into multiple random blocks of smaller size. Each block is processed independently through the semantic segmentation network, which prevents GPU memory overflow. The results are then aggregated to produce the final segmentation map for the entire image.
2Object-affected harmful factors
If remote sensing images are directly segmented into blocks, then GPU memory overflow is avoided, but the correlation within the image is lost
Solution Approach 1:
The patent processes multiple random blocks through the semantic segmentation network and then merges their results by aggregating the segmentation probability plots. This merging process preserves the spatial correlation and contextual information across the entire image while avoiding memory overflow during processing.
3Device complexity
If conventional semantic segmentation networks are used for multispectral remote sensing images, then the network structure is simple, but the accuracy of identifying objects with similar colors is insufficient
Solution Approach 1:
The patent applies different weighting coefficients to different spectral bands based on their specific characteristics and contributions to object identification. This localized quality adjustment allows the network to better distinguish objects with similar colors by emphasizing spectrally distinctive bands while downweighting redundant ones.
Data Source
AI summary
A semantic segmentation method and system for a high-resolution remote sensing image based on random blocks. In the semantic segmentation method, the high-resolution remote sensing image is divided into random blocks, and semantic segmentation is performed for each individual random block separately, thus avoiding overflow of GPU memory during semantic segmentation of the high-resolution remote sensing image. In addition, feature data in random blocks neighboring each random block incorporated into the process of semantic segmentation, overcoming the technical shortcoming that the existing segmentation method for the remote sensing image weakens the correlation within the image. Moreover, in the semantic segmentation method, semantic segmentation is separately performed on mono-spectral feature data in each band of the high-resolution remote sensing image, thus enhancing the accuracy of sematic segmentation of the high-resolution remote sensing image.