Crop-Row Image Segmentation With Minimal Training Labels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computer vision machine learning systems for agricultural robots require large amounts of labeled training data for supervised training, which is difficult to obtain, and unsupervised training is inadequate for high accuracy and wide generalizability in diverse agricultural environments.
Innovation Solution
A supervised training method for computer vision systems that uses a small dataset of labeled data, specifically 10 to 15 frames, to navigate robots along crop rows, with a human operator providing intuitive labeling inputs and photogrammetric analysis to generate labels, allowing the system to operate effectively in a single environment without extensive training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If supervised training is used to achieve high accuracy and wide generalizability, then navigation performance improves, but the amount of labeled training data required increases significantly
Solution Approach 1:
The system performs preliminary action by using photogrammetric analysis to automatically generate labeled training data before the supervised training process. This preliminary labeling step creates the necessary training dataset without requiring extensive manual annotation, enabling the system to achieve high navigation accuracy with minimal human intervention.
Solution Approach 2:
The system implements self-service by using its own captured images and photogrammetric capabilities to generate training labels autonomously. The robot captures images during navigation, and the photogrammetric analysis automatically produces labeled data, eliminating the need for external human annotators and reducing the practical burden of data collection.
2Measurement precision
If manually labeled training data is obtained through human annotators, then training data quality improves, but the time and resources required for data collection increase
Solution Approach 1:
The system performs self-service by automatically generating labels through photogrammetric analysis of images captured during robot navigation. This eliminates the need for human annotators to manually label images, significantly reducing data collection time while maintaining label accuracy through geometric computation.
Solution Approach 2:
The system replaces the mechanical process of manual human labeling with an automated photogrammetric analysis process. Instead of human annotators visually inspecting and labeling images, the system uses geometric principles and image processing algorithms to automatically generate accurate labels, thereby eliminating time loss associated with manual annotation.
3Reliability
If the system is trained on a single farm's crop rows, then navigation performance on that farm improves, but generalizability to other farms decreases
Solution Approach 1:
The system applies local quality by training the segmentation network specifically on images from a particular farm's crop rows, optimizing the model for that local environment. This localized training approach maximizes navigation accuracy on the specific farm while acknowledging that the model may require retraining for different farm environments.
Solution Approach 2:
The system enables parameter changes by allowing the segmentation network to be retrained with different farm-specific data. The model's parameters can be adjusted and retrained when deployed to new farms, transforming the system from a fixed generalizable model to an adaptable one that optimizes performance for each specific farm environment through parameter adjustment.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method enables accurate navigation of robots along crop rows with minimal training data, avoiding obstacles and adapting to specific agricultural environments, reducing the need for manual labeling and generalizing to other environments.
Implementation Method 1
segmentation networks, which are used to segment image data into one or more labeled segments
Implementation Method 2
conducting a photogrammetric analysis on the set of images to generate a set of label inputs on the set of images
Data Source
AI summary
Methods and systems related to computer vision for agricultural applications are disclosed herein. A disclosed method for navigating a robot along a crop row, in which each step is computer-implemented by a navigation system for the robot, includes capturing an image of at least a portion of the crop row, labeling, using a segmentation network, a portion of the image with a label, deriving a navigation path from the portion of the image and the label, generating a control signal for the autonomous navigation system to follow the navigation path, and navigating the robot along the crop row using the control signal.