Object Detection Model Training With Automated Defect Marking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
General users face challenges in using Low Code/No Code AI platforms for object detection model training due to insufficient training image quantity and quality, inconsistent marking, and the need for iterative data analysis, especially in factory environments, leading to inaccurate defect recognition.
Innovation Solution
A training system and method that includes a processing unit to iteratively train an object detection model, correct misjudgment sets, and adjust epoch numbers based on misjudgment values, ensuring the model avoids underfitting and overfitting, and outputs a master model for optimal performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the quantity of training images is increased to improve model accuracy, then the model performance is improved, but the time-consuming marking process and data collection difficulty increase
Solution Approach 1:
The system automatically generates training images by synthesizing defect information from unlabeled images, allowing the system to serve itself by creating training data without requiring manual marking for each image. The processing unit automatically identifies defect regions and generates corresponding training images with proper markings.
Solution Approach 2:
The system creates synthetic training images by copying and transforming defect information from source images. The processing unit extracts defect characteristics and generates new training images that replicate these patterns, providing sufficient training data without requiring original defect images for each training sample.
2Measurement precision
If manual marking of training images is performed to ensure data quality, then the marking precision is improved, but the marking consistency and efficiency deteriorate
Solution Approach 1:
The system performs automatic marking by having the processing unit identify defect regions and generate training images with proper annotations. This eliminates the need for manual marking while maintaining consistent and accurate labeling through algorithmic processing.
Solution Approach 2:
The patent replaces the mechanical manual marking process with an automated information processing system. The processing unit uses image processing algorithms to automatically identify defects and generate markings, substituting human labor with computational processes that are both efficient and consistent.
3Measurement precision
If iterative data analysis and model optimization are performed to improve defect recognition capability, then the model accuracy is improved, but the complexity of the training process and required data analysis ability increase
Solution Approach 1:
The system segments the complex training process into distinct functional modules: image collection, defect identification, training image generation, model training, and validation. Each module is handled by specific processing units, making the overall complex process more manageable and systematic.
Solution Approach 2:
The system incorporates feedback mechanisms where the processing unit analyzes model performance on validation sets and adjusts training parameters accordingly. The processing unit identifies misjudgment regions and uses this feedback to generate corrected training images, continuously optimizing model accuracy through automated feedback loops.
Data Source
AI summary
The present invention provides a training system and method, a testing system and method, a data filtering system and method, and a computer readable recording medium with stored program that includes verifying whether an object detection model has completed a training set learning and determining whether the object detection model is overfitting with a validation set, and outputting the training model as a master model before the object detection model is overfitting, and how to iterate the master model to match the test set test results and continuously maintain an online test level. The present invention provides a standardized method for achieving data marking consistency and training data selection.