Normalized Image Feature Extraction with Discriminative Loss
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition models struggle to accurately classify images due to the lack of effective methods for generating feature vectors with significant discrimination, leading to ambiguous similarity determinations between image categories.
Innovation Solution
A method involving a normalization processing module and a high discriminative loss function is introduced to guide the training of a feature extraction model, setting upper and lower limits for vector similarity based on image category similarities to enhance discrimination in feature vectors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image recognition models are used without specialized training methods, then the model structure remains simple, but the feature vectors lack significant discrimination and classification accuracy is poor
Solution Approach 1:
The patent applies parameter changes by introducing a high discriminative loss function with specific parameter constraints (upper limit and lower limit for vector similarity) to transform the training process. This loss function modifies the optimization parameters to ensure feature vectors achieve significant discrimination between categories, directly resolving the contradiction between classification accuracy and training method complexity.
Solution Approach 2:
The patent implements feedback through the high discriminative loss function that provides guidance during training. The loss function uses predefined similarity thresholds to feedback adjust the feature extraction model's output, ensuring that vector similarities meet the required discrimination criteria. This feedback mechanism improves classification accuracy while maintaining manageable training complexity.
2Reliability
If feature vectors are extracted without normalization processing, then the extraction process is faster, but the vector similarity calculations become ambiguous and less reliable
Solution Approach 1:
The patent applies preliminary action by incorporating normalization processing as a preliminary step before vector similarity calculation. This normalization is performed during the training process using the high discriminative loss function, which pre-adjusts the feature vectors to ensure reliable similarity determination. The preliminary normalization action eliminates ambiguity in subsequent similarity calculations while adding minimal training time overhead.
3Manufacturing precision
If the loss function does not enforce similarity limits, then the training process is simpler, but the feature vectors cannot achieve clear distinctions between categories
Solution Approach 1:
The patent uses parameter changes by defining specific parameter ranges (upper limit and lower limit) for vector similarity within the loss function. These parameter constraints transform the loss function to enforce clear distinctions between categories, achieving high precision in feature vector discrimination. The parameter-based approach maintains relative simplicity while significantly improving discrimination precision.
Data Source
AI summary
The present disclosure provides a method for training a feature extraction model, a method for classifying an image and related apparatuses, and relates to the field of artificial intelligence technology such as deep learning and image recognition. The scheme comprises: extracting an image feature of each sample image in a sample image set using a basic feature extraction module of an initial feature extraction model, to obtain an initial feature vector set; performing normalization processing on each initial feature vector in the initial feature vector set using a normalization processing module of the initial feature extraction model, to obtain each normalized feature vector; and guiding training for the initial feature extraction model through a preset high discriminative loss function, to obtain a target feature extraction model as a training result.