Fish School Simulation Images for ML Training Data Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in accurately estimating fish school information from captured images using machine learning models is the difficulty in creating a large amount of training data that covers various conditions, due to the manual effort required for assigning correct answer data, leading to suboptimal training quality and estimation accuracy.
Innovation Solution
An information processor generates simulation images of fish school behavior based on internal and external parameter values, including biological characteristics and shoaling parameters, to create high-quality training data for machine learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual assignment of correct answer data is used to create training data, then training data accuracy can be ensured, but the productivity of creating training data deteriorates due to the huge amount of manual work required
Solution Approach 1:
The patent uses simulation to generate synthetic training images that copy the essential characteristics of real fish school scenarios. By creating virtual images with controlled parameters (fish positions, densities, environments), the system produces large amounts of training data without manual annotation, while maintaining accuracy through programmatically generated ground truth labels.
Solution Approach 2:
The simulation system automatically generates both the training images and their corresponding correct answer data (ground truth) without human intervention. The system self-services the entire training data creation process by using parameterized models to produce consistent, accurate training pairs at scale.
2Measurement precision
If a large amount of training data covering various patterns is required to improve machine learning model accuracy, then the estimation accuracy of fish school information improves, but the device complexity and data management burden increase
Solution Approach 1:
The patent employs parameterized simulation models where key characteristics (fish school density, individual fish positions, environmental conditions) are controlled through adjustable parameters. By systematically varying these parameters, the system generates diverse training patterns without the complexity of manual data collection and management across multiple real-world scenarios.
Solution Approach 2:
The simulation system serves multiple functions: it generates training images, creates ground truth labels, varies environmental conditions, and adjusts fish behavior patterns all through a single unified platform. This multi-functionality reduces the overall system complexity compared to managing separate data collection and annotation processes.
Data Source
AI summary
An information processing method implemented by a computer. The information processing method includes the step of acquiring an internal parameter value regarding a biological characteristic of fish, an external parameter value regarding a surrounding environment characteristic of the fish, and a shoaling parameter value regarding a shoaling characteristic that is a behavior of one fish with respect to other fish, and the step of generating a simulation image including a behavior of each fish in a fish school based on the internal parameter value, the external parameter value, and the shoaling parameter value that are acquired in the step of acquiring.