Pixel-Level Crop Yield Prediction via Image Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current crop yield forecasting methods rely on manual and inaccurate techniques, with limited success from computer-based predictions, necessitating a system that can accurately predict crop yields in specific fields.
Innovation Solution
A yield prediction system that utilizes high-resolution aerial imagery and pixel-level analysis, incorporating an information gathering unit, analysis unit, and simulation unit to determine yield based on agronomic rules, stress levels, and image data, including noise removal techniques like erosion and blurring, to calculate predicted yields.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual methods and prior yields are used for crop yield forecasting, then the process is simple to implement, but the prediction accuracy is poor
Solution Approach 1:
The patent divides the field into multiple tiles and further segments each tile into individual pixels for analysis. This segmentation allows the system to process complex image data in manageable units, applying agronomic rules to each pixel to determine yield values. The segmented approach enables detailed pixel-level analysis without overwhelming computational complexity.
Solution Approach 2:
The patent transitions from traditional manual forecasting methods to a multi-dimensional approach by incorporating aerial imagery with multiple spectral channels (red, blue, green, NIR). This dimensional expansion allows the system to analyze crop characteristics across different spectral dimensions, significantly improving prediction accuracy while managing complexity through structured processing.
2Measurement precision
If pixel-level analysis with multiple spectral channels is performed, then the yield prediction accuracy improves, but the data processing complexity increases
Solution Approach 1:
The patent segments the complex multi-channel image data into manageable tiles, with each tile containing multiple spectral channels. This segmentation allows the system to process four-channel imagery (red, blue, green, NIR) in discrete units, applying agronomic rules to each pixel within each tile. The segmentation reduces the overall processing difficulty by breaking down the large-scale complex data into smaller, more manageable components.
Solution Approach 2:
The patent applies local quality by using erosion and blurring filters specifically on tile boundaries to remove noise. This targeted approach addresses processing difficulties locally at critical areas (tile edges) without requiring complex processing of the entire image, thereby improving measurement precision while managing overall processing difficulty.
3Measurement precision
If noise removal techniques like erosion and blurring are applied to tiles, then the measurement accuracy improves, but the processing time increases
Solution Approach 1:
The patent applies noise removal techniques selectively to tile boundaries rather than the entire image. By segmenting the processing scope to only the boundary regions where noise is most problematic, the system improves measurement precision at critical areas while minimizing the overall processing time required for noise removal.
Solution Approach 2:
The patent applies erosion and blurring filters locally at tile boundaries where noise removal is most needed for accurate pixel-level yield analysis. This localized application improves measurement precision where it matters most while avoiding the time penalty of applying the same processing to the entire image, thereby balancing accuracy and processing time.
Data Source
AI summary
A yield prediction system including an information gathering unit that retrieves a plurality of images of a field over a time period, an information analysis unit that divides each image into a plurality of tiles. a pixel analysis unit that gathers at least one agronomic rule to each tile and a simulation unit that determines the yield represented by each pixel in each image based on the agronomic rules and the analysis of each tile.