Individual Plant Recognition Using Image and Position Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies are unable to accurately recognize and distinguish between individual plants using digital images, especially when plants are in close proximity, due to insufficient accuracy of position coordinate sensors.
Innovation Solution
The use of machine learning models, such as convolutional neural networks (CNNs) trained with techniques like triplet loss, to recognize individual plants by analyzing digital images and additional data indicative of plant attributes, such as position coordinates, bounding shapes, and temporal data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If position coordinate sensors (GPS, IMU, triangulation-based sensors) are used to capture digital images of plants, then the system can obtain spatial information for plant identification, but the position coordinates are not sufficiently accurate to distinguish between different plants in close proximity
Solution Approach 1:
The system segments the plant identification problem into two parts: (1) using position coordinates for coarse localization and (2) using image processing to extract fine-grained plant-specific features. This segmentation allows the system to overcome the insufficient accuracy of position sensors by combining them with visual feature extraction from digital images.
Solution Approach 2:
The patent introduces digital image processing as an intermediary between position coordinate sensors and plant identification. The image processing extracts visual features (shape, color, texture, size) that serve as a mediator to bridge the gap between imprecise position data and accurate plant distinction, enabling identification when plants are in close proximity.
2Loss of information
If traditional plant identification technologies are used to classify plants based on image data, then plant types can be identified, but individual plants cannot be recognized as distinct from other individual plants
Solution Approach 1:
The system performs preliminary action by capturing multiple digital images of plants at different positions and angles before analysis. These pre-captured images contain various visual features that are later processed to extract unique plant identifiers, enabling individual plant recognition without requiring complex real-time processing during field operations.
Solution Approach 2:
The patent replaces traditional mechanical or manual plant identification methods with automated image processing and machine learning algorithms. This substitution enables the system to automatically extract and compare visual features from digital images, identifying individual plants based on their unique visual characteristics without human intervention.
3Loss of information
If multiple digital images of the same plant are captured at different positions and angles, then sufficient visual information can be obtained for identification, but the position coordinate accuracy remains insufficient to link images to specific plants
Solution Approach 1:
The system transitions from relying solely on the position coordinate dimension (which is imprecise) to utilizing the visual feature dimension extracted from digital images. By analyzing shape, color, texture, and size characteristics from multiple images, the system creates a new identification dimension that is independent of position coordinate precision, enabling accurate plant distinction despite spatial measurement limitations.
Data Source
AI summary
Implementations are described herein for training and applying machine learning models to digital images capturing plants, and to other data indicative of attributes of individual plants captured in the digital images, to recognize individual plants in distinction from other individual plants. In various implementations, a digital image that captures a first plant of a plurality of plants may be applied, along with additional data indicative of an additional attribute of the first plant observed when the digital image was taken, as input across a machine learning model to generate output. Based on the output, an association may be stored in memory, e.g., of a database, between the digital image that captures the first plant and one or more previously-captured digital images of the first plant.