Land Segmentation Model Using Semi-Supervised Learning for High-Precision Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current land classification methods face challenges in achieving high precision using low-precision training data, particularly in accurately segmenting and classifying land features like vegetation, which affects carbon sequestration calculations.
Innovation Solution
A method involving a segmentation model trained on diverse image data, including low-precision labels and unlabeled data, using convolutional neural networks (CNNs) with semi-supervised learning and preprocessing techniques to enhance accuracy, such as topographical normalization and up-sampling, to produce high-precision land segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If low-precision training data is used for land classification, then data availability and cost-effectiveness improve, but segmentation accuracy deteriorates
Solution Approach 1:
The patent applies preliminary action by performing data augmentation and preprocessing operations on low-precision training data before model training. Techniques such as geometric transformations, color space conversions, and synthetic data generation are performed in advance to enhance the quality and variability of training samples, thereby improving segmentation accuracy without requiring high-precision original data
Solution Approach 2:
The patent employs parameter changes by transforming the training data through various parameter modifications including resolution adjustments, color space conversions (RGB to LAB, HSV), and geometric parameter variations. These parameter transformations enable the model to learn robust features from low-precision data while achieving high segmentation accuracy
2Productivity
If simple classification models are used, then processing speed improves, but classification precision deteriorates
Solution Approach 1:
The patent applies segmentation by dividing the land classification task into multiple specialized sub-tasks handled by different network branches. The multi-task learning architecture segments the processing into vegetation detection, building detection, and land use classification, allowing each branch to specialize and achieve high precision while maintaining efficient parallel processing
Solution Approach 2:
The patent implements universality through a multi-functional neural network architecture that simultaneously performs multiple classification tasks (vegetation, building, land use) within a single model. This universal model processes different land features in parallel, achieving both high classification precision across multiple categories and efficient processing speed through unified architecture
Data Source
AI summary
A method for segmenting and classifying a unit of land, comprising: receiving land information comprising image data; processing received land information using a segmentation model trained on training data comprising training image data; and determining segmentation and classification for the unit of land; wherein the training image data comprises image data from different time points; and wherein the training image data comprises at least some of the image data.