AI Image Segmentation With Continuous Model Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated optical inspection (AOI) systems in factories face inefficiencies due to labor-intensive and costly data collection, management, and maintenance of AI models, requiring manual adjustments and interruptions for model retraining, which affects detection quality and efficiency.
Innovation Solution
A detecting system and method that utilizes a camera and processor to capture and process images, automatically training AI models using multiple images of different target objects, allowing simultaneous detection and model training without interruption, reducing reliance on human operators and improving detection efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual data collection and model retraining are performed, then detection quality is improved, but detection efficiency deteriorates due to interruptions and labor-intensive processes
Solution Approach 1:
The system enables the AI model to train itself automatically using captured images without requiring manual data collection, marking, or model retraining. The processor automatically performs model training in the background while detection operations continue, making the system self-sufficient and eliminating dependency on human operators for model maintenance.
Solution Approach 2:
The system maintains continuous detection operations while simultaneously performing model training in the background. The processor executes detection tasks without interruption and conducts model training using captured images during idle processing time, ensuring that useful actions (detection and model improvement) continue uninterrupted and continuously.
2Measurement precision
If automated optical inspection devices are equipped with AI technology, then detection quality is improved, but operational complexity increases due to data management and model maintenance requirements
Solution Approach 1:
The AI model performs self-training automatically using images captured by the camera. The processor automatically manages data collection, preprocessing, and model retraining without requiring human intervention for data marking or model maintenance, significantly simplifying operational complexity while maintaining improved detection quality.
Solution Approach 2:
The system merges the detection function and model training function into a single integrated workflow. Both operations share the same hardware resources (camera and processor) and operate simultaneously, eliminating the need for separate data collection and model training processes, thereby reducing operational complexity.
3Measurement precision
If model retraining is performed manually, then detection quality is maintained, but time consumption increases due to labor-intensive data marking and management
Solution Approach 1:
The system automatically performs model retraining using captured images without requiring manual data marking or management. The processor autonomously handles the entire model training process, including data preprocessing and parameter updates, eliminating the time-consuming manual operations while maintaining detection quality.
Solution Approach 2:
The system performs preliminary data preparation by capturing and storing images during normal detection operations. These pre-captured images are readily available for immediate model training when needed, eliminating the time required for on-demand data collection and marking, thus enabling rapid model retraining while maintaining quality.
Data Source
AI summary
A detecting system includes a camera and a processor. The camera is configured to capture a first target object to generate a first image and is configured to capture a second target object different from the first target object to generate a second image. The processor is configured to detect the first image by using a first model to generate a first result and is configured to train the first model by using the first result. When the processor trains the first model by using the first result, the camera captures the second target object. After the camera captures the second target object, the processor further trains the first model by using the second image.