Crop Field Boundary Detection via ML Enhancement Channel
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques face challenges in accurately segmenting distinct crop fields from each other and surrounding areas using high-elevation imagery, often resulting in blurred boundaries between neighboring fields of different crops.
Innovation Solution
The method involves generating a 'boundary enhancement channel' using spectral channels of high-elevation imagery as input across machine learning models, specifically convolutional neural networks (CNNs), to accurately detect boundaries between crop fields and other areas, with pixel-wise classifications of angles to a reference location such as the center of mass or nearest boundary, and processing this channel to determine pixel-wise field memberships.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image segmentation techniques are used on satellite data, then processing speed and simplicity are maintained, but boundary identification accuracy between fields deteriorates
Solution Approach 1:
The patent segments the image processing task into two distinct components: a boundary detection network that identifies field boundaries, and a field segmentation network that assigns pixels to fields. This segmentation allows each network to specialize in one function, improving boundary identification accuracy while maintaining computational efficiency through division of labor.
Solution Approach 2:
The patent introduces an intermediary boundary detection network that generates boundary probability maps, which then serve as guidance for the field segmentation network. This intermediary component acts as a mediator that translates raw image data into structured boundary information, significantly improving boundary identification accuracy without requiring the main segmentation network to handle both boundary detection and field assignment simultaneously.
2Measurement precision
If machine learning models with atrous convolution and ASPP are used, then boundary detection precision is improved, but computational complexity and processing time increase
Solution Approach 1:
The patent applies atrous convolution and ASPP modules selectively only in the boundary detection network where they are most needed for precise boundary identification, rather than applying them throughout the entire segmentation pipeline. This partial application maintains high boundary detection precision while minimizing the overall computational burden and processing time.
Solution Approach 2:
By separating boundary detection from field segmentation into two independent networks, the patent allows complex operations like atrous convolution and ASPP to be confined to the boundary detection network. This segmentation enables optimized resource allocation where computational intensity is concentrated only where needed, reducing overall processing time while maintaining precision.
3Measurement precision
If pixel-wise crop type classification is performed first, then crop classification accuracy is improved, but the complexity of the processing pipeline increases
Solution Approach 1:
The patent performs pixel-wise crop type classification as a preliminary step before field segmentation. By pre-classifying each pixel's crop type, the system establishes accurate crop type labels that can be directly assigned to segmented fields, improving crop classification accuracy while keeping the overall pipeline manageable through clear sequential stages.
Solution Approach 2:
The patent merges the crop type classification results with the field segmentation output by assigning the pre-determined crop type labels to each segmented field. This merging approach integrates two separate functions (classification and segmentation) into a unified workflow, improving overall accuracy while avoiding the need for a single overly complex model.
Data Source
AI summary
Implementations relate to improved crop field segmentation and crop classification in which boundaries between crop fields are more accurately detected. In various implementations, high-elevation image(s) that capture an area containing multiple demarcated fields may be applied as input across one or more machine learning models to generate a boundary enhancement channel. Each pixel of the boundary enhancement channel may be spatially aligned with a corresponding pixel of the one or more high-elevation images. Moreover, each pixel of the boundary enhancement channel may be classified with a unit angle to a reference location of the field of the multiple demarcated fields that contains the pixel. Based on the boundary enhancement channel, pixel-wise field memberships of pixels of the one or more high-elevation images in the multiple demarcated fields may be determined.