Deep Learning Land Cover Inference With HRLC Training Layers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing satellite image-based land cover prediction methods require significant data and computational resources, and there is a need for more efficient and accurate machine learning models that can produce high-resolution land cover classifications.
Innovation Solution
A deep learning-based method using a trained model that utilizes a subset of independent variables from High-Resolution Land Cover (HRLC) data to generate high-resolution land cover predictions, incorporating techniques like convolutional neural networks and post-classification ruleset corrections to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning models use significant data and many independent variables to produce accurate high-resolution land cover predictions, then prediction accuracy is improved, but computational resource requirements and data processing complexity increase
Solution Approach 1:
The patent extracts and utilizes pre-generated HRLC final land cover layers as training data, removing the need to process raw satellite images and multiple independent variables from scratch. This extraction of refined data reduces computational complexity while maintaining prediction accuracy.
Solution Approach 2:
The patent performs preliminary land cover classification to generate HRLC final layers before training the deep learning model. This preliminary action prepares refined training data in advance, reducing the computational burden during model training and inference while preserving high prediction accuracy.
2Measurement precision
If traditional methods process multiple satellite images and perform numerous transformations to generate high-resolution land cover maps, then classification accuracy is improved, but processing time and computational resources increase
Solution Approach 1:
The patent performs image transformations and temporal stack layer creation in advance during the training phase. By preparing these processed layers beforehand, the model can quickly infer land cover classifications without performing repeated transformations during actual classification tasks, significantly reducing processing time.
Solution Approach 2:
The patent creates temporal stack layers that capture temporal statistics as reusable data structures. These copied and standardized data representations can be efficiently processed by the deep learning model without requiring repeated complex transformations of original satellite images.
3Adaptability or versatility
If deep learning models are trained on diverse geographic regions, then model adaptability and versatility are improved, but training data requirements and computational resources increase
Solution Approach 1:
The patent trains the deep learning model on HRLC final land cover layers from multiple diverse geographic regions (California, Idaho, Montana, Washington). This multi-regional training creates a universal model that can accurately classify land cover across different geographic locations without requiring region-specific training data, enhancing model versatility.
Solution Approach 2:
The patent uses pre-generated HRLC final layers from different regions as training samples, copying the essential land cover patterns and characteristics across regions. This approach allows the model to learn universal land cover classification patterns without processing raw satellite data from each region, reducing overall training data requirements.
Data Source
AI summary
A method for performing land classification operations, the method can comprise receiving, by one or more servers, a plurality of High Resolution Land Cover (“HRLC”) final land cover layers; training, by one or more servers, a model using the plurality of HRLC final land cover layers to form a trained deep learning HRLC (“DL-HRLC”) model; and transferring, by the one or more servers, the trained DL-HRLC model for land cover prediction, wherein a land cover inference engine of the trained DL-HRLC model classifies a single image, which comprises a plurality of pixels, to generate an output land cover layer prediction.