Few-shot urban remote sensing image information extraction method based on meta learning and attention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing few-shot urban remote sensing image information extraction methods face challenges due to weak generalization ability and require large-scale training samples, making it difficult to achieve accurate extraction of urban ground features.
Innovation Solution
A method combining meta learning and attention mechanisms to build a pre-trained model that focuses on 'time-space-spectrum-angle' information, suppressing noise and leveraging parallel association learning to optimize feature parameters for efficient extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning algorithm is used for high resolution image information extraction, then image recognition ability is improved, but large-scale training samples and parameters are required which increases cost and complexity
Solution Approach 1:
The patent applies pre-training on large-scale remote sensing images to initialize the network parameters before fine-tuning on small-scale urban regions. This preliminary action allows the model to learn general features from abundant data, then adapt to specific few-shot scenarios without requiring large-scale labeled training for each region, thus resolving the contradiction between high recognition ability and training sample requirements
Solution Approach 2:
The patent segments the training process into two distinct phases: pre-training phase on large-scale remote sensing images to learn general features, and fine-tuning phase on small-scale urban regions to adapt to specific scenarios. This segmentation allows the model to benefit from large-scale data without requiring large-scale labeled data for each specific application, reducing the complexity of training sample collection
2Measurement precision
If large-scale training sample labeling is performed for deep learning, then extraction accuracy is improved, but cost and time consumption increase due to geological regions and weather limitations
Solution Approach 1:
The patent performs preliminary pre-training on large-scale remote sensing images with general features before fine-tuning on small-scale labeled urban data. This preliminary action allows the model to acquire robust feature extraction capabilities from abundant unlabeled or lightly-labeled data, reducing the need for extensive manual labeling of training samples for each specific region and time period
Solution Approach 2:
The patent uses pre-trained model parameters as a foundation that can be copied and adapted to different urban regions and time periods. Instead of creating new training samples and retraining from scratch for each scenario, the model copies the pre-learned features and fine-tunes them on small labeled datasets, significantly reducing labeling time and cost while maintaining extraction accuracy
3Productivity
If deep learning algorithm is applied to few-shot urban regions, then information extraction is performed, but generalization ability becomes weak leading to inaccurate extraction
Solution Approach 1:
The patent applies pre-training on diverse large-scale remote sensing images before fine-tuning on few-shot urban regions. This preliminary action enables the model to learn robust and transferable features that generalize well to unseen regions. The pre-trained model serves as a strong foundation that compensates for the limited data in few-shot scenarios, improving both extraction capability and generalization reliability
Solution Approach 2:
The patent creates a universal pre-trained model that can be applied to multiple different urban regions and time periods. The model learns general remote sensing features that are transferable across different scenarios, making it universally applicable to few-shot regions without requiring region-specific large-scale training data, thus improving both productivity and reliability
Data Source
AI summary
A few-shot urban remote sensing image information extraction method based on meta learning and attention includes building a few-shot urban remote sensing information pre-trained model. During a pre-training stage, pre-training network learning is performed for a few-shot set to fully learn feature information of existing samples and obtain initial feature parameters and a deep convolutional network backbone of the few-shot set; the few-shot urban remote sensing information pre-trained model is a network structure including a convolutional layer, a pooling layer and a fully-connected layer, and includes five sections of convolutional network where each section includes two or three convolutional layers, and an end of each section is connected to one maximum pooling layer to reduce a size of a picture; the number of convolutional kernels inside each section is same, and when closer to the fully-connected layer, the number of convolutional kernels is larger.