Dual Framework Model Training for Degraded Image Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image recognition performance significantly deteriorates when images are occluded, in different postures, fuzzy, low resolution, or of poor quality, as existing technologies fail to effectively handle variations in image data.
Innovation Solution
An apparatus and method for training a classifying model using two machine learning frameworks with identical structures and parameters, where one framework processes sample images and the other processes corresponding images with variations, calculating a similarity loss to optimize the total loss and adjust parameters, thereby improving recognition performance for degraded images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a single machine learning framework is used for image classification, then the training process is simple, but the recognition performance deteriorates significantly when images are occluded, in different postures, fuzzy, low resolution, or poor quality
Solution Approach 1:
The patent merges two machine learning frameworks into a unified dual-framework system. The first framework processes sample images while the second framework processes corresponding degraded images (occluded, fuzzy, low-resolution, etc.). Both frameworks share parameters and are trained simultaneously with a combined loss function, enabling the system to maintain high recognition performance across various image quality conditions without excessive complexity increase.
Solution Approach 2:
The dual machine learning framework system achieves multi-functionality by handling multiple types of image degradation simultaneously. The same framework structure can process clean images, occluded images, fuzzy images, low-resolution images, and poor-quality images, making the system universally applicable to various real-world imaging scenarios while maintaining consistent performance.
2Adaptability or versatility
If the model is trained only on clear sample images, then the training data requirement is simple, but the model cannot handle degraded images effectively
Solution Approach 1:
The patent applies preliminary action by pre-processing sample images to generate corresponding degraded images through simulated occlusion, blurring, downsampling, and quality degradation. These pre-generated degraded images serve as training data for the second framework, enabling the model to learn degradation patterns and maintain performance without requiring large quantities of actual degraded training images.
Solution Approach 2:
The system creates copies of original sample images and applies various degradation transformations to generate synthetic degraded versions. These copied and transformed images are then used to train the second framework, allowing the model to generalize to real degraded images without needing extensive collections of actual degraded training data.
Data Source
AI summary
An apparatus for training a classifying model comprises: a first obtaining unit configured to input a sample image to a first machine learning framework, to obtain a first classification probability and a first classification loss; a second obtaining unit configured to input a second image to a second machine learning framework, to obtain a second classification probability and a second classification loss, the two machine learning frameworks having identical structures and sharing identical parameters; a similarity loss calculating unit configured to calculate a similarity loss related to a similarity between the first classification probability and the second classification probability; a total loss calculating unit configured to calculate the sum of the similarity loss, the first classification loss and the second classification loss, as a total loss; and a training unit configured to adjust parameters of the two machine learning frameworks to obtain a trained classifying model.