Remote Sensing Image Recognition via Segmentation and Resolution Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolutional neural networks in remote sensing image recognition often suffer from limited receptive fields, leading to misrecognition due to insufficient environmental information, and result in high computational costs.
Innovation Solution
The method involves reducing the resolution of remote sensing images, segmenting them into blocks, selecting blocks based on pixel confidence, and processing these blocks through a neural network to enhance receptive fields and reduce computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the receptive field of a neuron in a convolutional neural network is limited, then the calculation amount is reduced, but the recognition accuracy deteriorates due to insufficient environmental information
Solution Approach 1:
The patent divides the remote sensing image into multiple image blocks and processes them separately through the convolutional neural network. Each image block is fed into the network independently, allowing the neuron's receptive field to effectively cover the entire block while maintaining manageable computational complexity. This segmentation approach resolves the contradiction by making the effective receptive field size proportional to the input block size rather than the entire image size.
Solution Approach 2:
The patent introduces a spatial dimension by dividing the image into blocks and processing them in parallel. Instead of attempting to process the entire large image with a single neural network pass (which would require a prohibitively large receptive field), the system processes multiple smaller blocks simultaneously, effectively utilizing the limited receptive field capacity across multiple spatial regions.
2Productivity
If the resolution of the remote sensing image is reduced, then the computational load is reduced, but the recognition accuracy may deteriorate due to loss of detail
Solution Approach 1:
The patent segments the image into blocks and processes each block at a reduced resolution through the neural network. By dividing the image, the system can maintain higher effective resolution within each block while reducing the overall computational load. The segmentation allows the network to process smaller, more manageable units at lower resolution without losing the ability to accurately identify objects within those units.
3Measurement precision
If the entire remote sensing image is processed through the neural network, then the recognition accuracy is improved, but the calculation amount increases greatly
Solution Approach 1:
The patent divides the remote sensing image into multiple smaller image blocks and processes each block independently through the neural network. This segmentation reduces the computational complexity significantly compared to processing the entire image at once, as the network only needs to process each block's pixels rather than all pixels in the large image. The accuracy is maintained because each block is processed with sufficient detail for its size.
Solution Approach 2:
Instead of processing the entire image (excessive action), the patent processes only the necessary portions (image blocks) that contain the objects of interest. This partial processing approach reduces the total calculation amount while maintaining adequate accuracy for the identified objects within each block.
Data Source
AI summary
A remote sensing image recognition method, apparatus and device, a storage medium and a computer program. The remote sensing image recognition method comprises: performing resolution reduction processing on a remote sensing image to be recognized to obtain a resolution reduction processed remote sensing image (S100); segmenting at least one remote sensing image block from the resolution reduction processed remote sensing image (S110); determining, from the at least one remote sensing image block, a remote sensing image block to be processed (S120); inputting, into a neural network, the remote sensing image block to be processed, and obtaining, through the neural network, classification probability information of pixels in the input remote sensing image block to be processed (S130); and determining a recognition result of the remote sensing image to be recognized, according to the classification probability information of the pixels in the remote sensing image block to be processed (S140).