Aquaculture Underwater Camera Positioning with Reinforcement Learning

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The precise positioning of underwater cameras in aquaculture environments is crucial for accurate biomass estimation and sea lice counting, as improper placement can lead to low-quality images and unreliable computer vision results, impacting the effectiveness of fish farming operations and compliance with regulations.

Innovation Solution

A reinforcement learning approach is employed to balance exploitation of known favorable camera positions with exploration of new positions, using data from previous actions to formulate a camera position policy that maximizes outcomes over time.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the camera is positioned far from the aquatic organisms, then the camera can be placed in a stable location, but the image quality deteriorates and computer vision accuracy decreases

Engineering Contradiction:

Improvecamera positioning stabilityVSAvoidbiomass estimation accuracy

Core Design Contradiction:

ReliabilityVSMeasurement precision

Solution Approach 1:

The system transitions from static camera positioning to dynamic positioning by implementing a reinforcement learning agent that continuously adjusts camera positions based on real-time feedback from computer vision performance metrics, enabling the camera to adapt its location to maintain optimal image quality while accounting for moving aquatic organisms

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system implements a closed-loop feedback mechanism where computer vision results (biomass estimation accuracy, sea lice counting precision) are fed back to the reinforcement learning agent, which then adjusts camera positioning decisions to improve future image quality and measurement accuracy

Inventive Principle:

Principle #23Feedback

2Measurement precision

If the camera is positioned close to the aquatic organisms, then the image quality and computer vision accuracy improve, but the camera may miss broader population patterns

Engineering Contradiction:

Improvesea lice counting accuracyVSAvoidpopulation-wide monitoring capability

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The reinforcement learning agent dynamically adjusts camera positioning between close-range and wide-range positions based on real-time assessment of what is most beneficial for current monitoring objectives, allowing the system to adaptively switch between detailed inspection mode and population survey mode

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system changes the spatial parameter (camera position) based on reinforcement learning decisions, transitioning between different distance ranges to optimize for either detailed organism inspection or broader population pattern detection depending on current needs

Inventive Principle:

Principle #35Parameter changes

3Ease of operation

If fixed camera positions are used, then the system is simple to operate, but the system cannot adapt to varying environmental conditions and organism behavior

Engineering Contradiction:

Improvecamera positioning simplicityVSAvoidenvironmental condition adaptation

Core Design Contradiction:

Ease of operationVSAdaptability or versatility

Solution Approach 1:

The system implements self-service by enabling the camera positioning system to automatically adapt to environmental conditions and organism behavior through reinforcement learning, eliminating the need for manual intervention while maintaining operational simplicity for the end user

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system transitions from static fixed positions to dynamic adaptive positioning where the reinforcement learning agent continuously learns from environmental feedback and adjusts camera positions autonomously to maintain optimal performance across varying conditions

Inventive Principle:

Principle #15Dynamics

4Measurement precision

If the camera frequently changes position to track organisms, then image quality is maintained, but the system complexity and energy consumption increase

Engineering Contradiction:

Improvebiomass estimation accuracyVSAvoidcamera positioning energy consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The reinforcement learning agent implements partial action by selectively adjusting camera positions only when and where needed to maintain adequate image quality, rather than continuously tracking every organism movement, thus reducing unnecessary energy consumption while preserving measurement accuracy

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The system optimizes the position parameter by making targeted adjustments based on reinforcement learning decisions rather than continuous changes, reducing the frequency and magnitude of camera movements to minimize energy consumption while maintaining sufficient image quality for accurate biomass estimation

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12418721B2Position policy determination for underwater camera positioning in an aquaculture environment

Publication Date: 2025.09.16 AQUABYTE INC

AI summary

A system and method involve determining a policy for positioning an underwater camera in an aquaculture environment. The system and method include receiving data on previous camera positioning actions and results, using this data to determine a new camera positioning action policy, receiving a request to choose a camera positioning action based on the new policy, determining a specific camera positioning action based on the policy, and providing the chosen action in response to the request.