Machine Learning Fish Re-Identification Using Spot Pattern Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
It is difficult to discern between different fish of the same species, particularly in aquaculture settings, which hinders accurate tracking, disease detection, and pre-sale estimation of fish populations.
Innovation Solution
A machine learning model is trained to identify and re-identify fish based on their unique spot patterns, generating embeddings that are mapped in a high-dimensional space for comparison, allowing for precise tracking and re-identification of individual fish.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional visual inspection methods are used to identify fish, then the process is simple and requires no complex equipment, but it is impossible to discern between different fish of the same species
Solution Approach 1:
The patent replaces manual visual inspection with an automated machine learning system that uses image processing and neural networks to identify fish. The system captures images of fish, extracts features such as spot patterns and body shape, and uses trained models to automatically distinguish between individual fish of the same species, eliminating the need for human visual discrimination while achieving high identification accuracy.
Solution Approach 2:
The system creates digital representations (embeddings) of fish based on their visual characteristics. These embeddings are mathematical vectors that capture the unique features of each fish, allowing the system to store and compare fish identities without physically handling or marking the fish. This digital copying approach enables precise identification while maintaining system simplicity.
2Reliability
If fish are tracked individually in large-scale aquaculture operations, then disease detection and health monitoring improve, but the ability to operate in real ocean conditions with 200,000+ fish becomes limited
Solution Approach 1:
The machine learning system serves multiple functions simultaneously: it identifies individual fish, tracks their movement, monitors their health conditions, and detects diseases. By consolidating these functions into a single automated system, the patent enables effective individual tracking even in large-scale operations with 200,000+ fish, overcoming the limitation of manual monitoring methods.
Solution Approach 2:
The system automatically processes and analyzes fish images without requiring human intervention. The machine learning models self-adjust and improve through continuous learning from new data, enabling the system to handle large volumes of fish independently. This automation allows the system to scale to large populations while maintaining reliable disease detection and health monitoring capabilities.
3Measurement precision
If fish spot patterns are used for identification, then unique and identifiable characteristics can be tracked over time, but spots may grow in size and new spots may develop
Solution Approach 1:
The system divides the fish identification task into multiple independent feature extraction components. Instead of relying solely on spot patterns, the system segments the analysis into multiple features including spot location, size, shape, body contour, and other morphological characteristics. This segmentation allows the system to compensate for changes in individual spot patterns while maintaining overall identification accuracy.
Solution Approach 2:
The identification system uses a composite approach by combining multiple features into a unified embedding representation. The final fish identity is determined by synthesizing information from various features (spot patterns, body shape, color distribution, etc.), creating a robust identification system that remains accurate even when individual features change over time. This composite approach is analogous to using composite materials that maintain structural integrity despite changes in individual components.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on computer storage media for identification and re-identification of fish. In some implementations, first media representative of aquatic cargo is received. Second media based on the first media is generated, wherein a resolution of the second media is higher than a resolution of the first media. A cropped representation of the second media is generated. The cropped representation is provided to the machine learning model. In response to providing the cropped representation to the machine learning model, an embedding representing the cropped representation is generated using the machine learning model. The embedding is mapped to a high dimensional space. Data identifying the aquatic cargo is provided to a database, wherein the data identifying the aquatic cargo comprises an identifier of the aquatic cargo, the embedding, and a mapped region of the high dimensional space.