Sticker Image Classification Across Far-Edge and Near-Edge AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional AI workflows face difficulties in processing large images from large manufacturing systems in real time, necessitating improved edge AI applications for efficient sticker detection on industrial products.
Innovation Solution
A hierarchical edge-computing architecture involving far edge, near edge, and cloud resources is employed to decouple deep learning models, enabling efficient sticker detection and retraining, reducing manual workload through incremental learning and model flexibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional AI workflow is used for surface defect detection, then model recognition accuracy can be improved with advanced deep learning technology, but real-time processing of large images from manufacturing systems becomes difficult
Solution Approach 1:
The patent segments the image processing workflow into two distinct parts: (1) far-edge devices perform lightweight object detection to locate regions of interest (stickers) and (2) near-edge devices perform comprehensive classification on the extracted sticker images. This segmentation allows the system to maintain high accuracy through specialized classification while achieving real-time performance through efficient region extraction at the far edge.
Solution Approach 2:
The patent introduces an intermediary mechanism where the far-edge device extracts and transmits only the relevant sticker regions to the near-edge device for classification. This intermediary step filters out unnecessary background information, reducing the computational burden on the classification model while maintaining high detection accuracy for the actual defect areas.
2Productivity
If a hierarchical far edge-near edge-cloud architecture is implemented with decoupled deep learning models, then workload can be balanced and detection efficiency improved, but system complexity increases
Solution Approach 1:
The system segments computational tasks across three hierarchical levels: far-edge devices handle object detection and region extraction, near-edge devices perform sticker classification, and cloud devices conduct model retraining. This segmentation distributes workload appropriately, improving overall detection efficiency while managing complexity through clear functional separation at each level.
Solution Approach 2:
The patent implements multi-functionality at different hierarchical levels: far-edge devices perform both object detection and region extraction, near-edge devices handle classification and can receive retrained models, and cloud devices perform model retraining and deployment. This universal design allows each component to handle multiple related tasks, improving efficiency while maintaining manageable complexity through standardized interfaces.
3Speed
If the far edge extracts regions of interest and sends them to the near edge for classification, then processing speed improves, but data transmission requirements increase
Solution Approach 1:
The far-edge device extracts only the relevant sticker regions from the full product images and transmits only these extracted regions to the near-edge device for classification. This extraction process removes unnecessary background data, significantly reducing the volume of transmitted data while maintaining high processing speed by focusing computational resources on the actual defect areas.
Data Source
AI summary
Embodiments of the present disclosure relate to a method, a device, and a computer program product for detecting stickers. The method in an illustrative embodiment includes sending, at a far edge, a sticker image of an object device to a near edge, wherein a sticker is attached to a surface of the object device, and the sticker image comprises an image of the sticker. The method further includes querying, at the far edge, label information of the object device. The method further includes receiving, at the far edge from the near edge, a classification result from classifying the sticker image. The method further includes comparing, at the far edge, the classification result with the label information. The method further includes sending, from the far edge, the sticker image to a cloud via the near edge in response to determining that the classification result is not matched with the label information.