Agricultural Field Boundary Detection Using Satellite Imagery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining agricultural field boundaries are time-consuming and costly, especially for large scales, as they often require manual annotation or data collection from decentralized offices, making it difficult to obtain and utilize accurate geospatial data for precision farming and landscape mapping.
Innovation Solution
A computer-implemented method using optical satellite data to classify boundary elements from primary images, constructing a polygon-based representation of agricultural field boundaries, which can be used for geospatial applications such as controlling agricultural machinery, and integrating trained artificial neural networks for classification and boundary detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation or data collection from decentralized offices is used to determine field boundaries, then measurement precision is improved, but productivity deteriorates due to time-consuming and costly processes
Solution Approach 1:
The patent replaces manual annotation and physical data collection processes with an automated computer vision system that processes satellite imagery. The system uses trained neural networks to automatically detect and classify boundary elements in satellite images, eliminating the need for manual field measurements and data entry while maintaining high precision in boundary determination.
Solution Approach 2:
The patent creates a digital representation (polygon-based format) of physical field boundaries by processing satellite image data. This digital copy accurately represents the actual field boundaries without requiring physical measurement, enabling rapid replication and use of boundary data across multiple applications and regions.
2Measurement precision
If manual methods are used for large-scale field boundary determination, then measurement precision is maintained, but loss of time increases significantly
Solution Approach 1:
The patent employs pre-trained neural networks that have been previously trained on labeled satellite imagery. These pre-trained models can immediately process new satellite images to detect boundaries without requiring time-consuming training for each new region, significantly reducing processing time while maintaining consistent precision across different geographic areas.
Solution Approach 2:
The automated image processing system replaces slow manual annotation processes with parallel computation capabilities that can analyze multiple satellite images simultaneously. The computer vision pipeline processes entire regions efficiently by automatically detecting boundary elements and constructing polygon representations without human intervention.
3Productivity
If satellite data is used for boundary determination, then productivity is improved through automated processing, but measurement precision may deteriorate without proper training data
Solution Approach 1:
The system performs preliminary training of neural networks using labeled satellite imagery containing known boundary information. This pre-training phase enables the model to learn accurate boundary detection patterns before being deployed for automated processing, ensuring high precision is achieved before productivity benefits are realized.
Solution Approach 2:
The system uses labeled satellite imagery as ground truth feedback to train and validate the neural network models. The labeled data provides continuous feedback on detection accuracy, allowing the system to refine its boundary detection algorithms and maintain high precision as automation scales to larger regions.
Data Source
Figure 1~4b
Figure 2~4a
Figure 3a~3c
AI summary
The invention relates to the field of evaluating an agricultural field, particularly to a computer-implemented method for determining a boundary (75) of the at least one agricultural field (70) for controlling geospatial applications. The method comprises the steps of: obtaining a plurality of optical satellite data as primary images (35), the primary images (35) including a representation of the at least one agricultural field (70); consolidating the plurality of primary images (35) to an intermediate image (45), the intermediate image (45) comprising a plurality of elements (47); classifying each element (47) of the intermediate image (45) as a boundary element and/or as a non boundary element; and constructing a representation of the boundary (65) of the agricultural field (70) in a polygon-based format, based on connecting at least a subset of the boundary elements, as a basis for geospatial applications, particularly for controlling agricultural machinery.