Regional Proposal Network for Label Classification in Mixed-Quality Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional AI-based approaches struggle to accurately classify labels in non-studio images of products due to inferior lighting, focus, framing, resolution, and background conditions, resulting in low accuracy and efficiency.
Innovation Solution
The use of a regional-based AI model, specifically a regional-based convolutional neural network (R-CNN), which includes a regional proposal network (RPN) to detect labels of interest and a CNN to classify labels, focusing on regions of interest and discarding irrelevant regions to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional AI-based approaches are used to classify labels in non-studio images, then the system can process a high volume of images, but the classification accuracy deteriorates due to inferior lighting, focus, framing, resolution, and background conditions
Solution Approach 1:
The patent applies segmentation by dividing the image processing task into two distinct stages: (1) a regional proposal network that identifies and segments regions of interest containing potential labels, and (2) a CNN that classifies labels only within these segmented regions. This segmentation allows the system to process high volumes of images while maintaining accuracy by focusing computational resources on relevant areas rather than entire images, thereby resolving the contradiction between productivity and measurement precision.
Solution Approach 2:
The patent implements local quality by applying different processing strategies to different parts of the image. The regional proposal network identifies regions with varying quality characteristics (some regions may have poor lighting or focus), and the system applies targeted processing only to those regions. This allows the system to maintain high classification accuracy for labels in challenging conditions while efficiently processing the overall high volume of images, thus resolving the contradiction between productivity and measurement precision.
2Adaptability or versatility
If traditional AI models process entire images to classify labels, then they can handle diverse image conditions, but the processing efficiency deteriorates due to computational complexity
Solution Approach 1:
The patent applies the extraction principle by removing irrelevant regions from the processing pipeline. The regional proposal network extracts only the portions of images that contain potential labels, discarding the majority of the image data that does not contribute to label classification. This extraction maintains adaptability to diverse image conditions by still processing various image types, while dramatically improving processing efficiency by reducing the computational burden, thus resolving the contradiction between adaptability and productivity.
Solution Approach 2:
The patent implements preliminary action by performing region detection and filtering before the actual label classification. The regional proposal network预先 identifies and prepares regions of interest, so that when the CNN performs classification, it only needs to process pre-selected regions rather than entire images. This preliminary action maintains versatility in handling diverse image conditions while improving processing efficiency, resolving the contradiction between adaptability and productivity.
Data Source
AI summary
Example methods, apparatus, and articles of manufacture to classify labels based on images using artificial intelligence are disclosed. An example apparatus includes a regional proposal network to determine a first bounding box for a first region of interest in a first input image of a product; and determine a second bounding box for a second region of interest in a second input image of the product; a neural network to: generate a first classification for a first label in the first input image using the first bounding box; and generate a second classification for a second label in the second input image using the second bounding box; a comparator to determine that the first input image and the second input image correspond to a same product; and a report generator to link the first classification and the second classification to the product.