Fine-Grained Image Recognition via Grouped Feature Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Fine-grained image recognition faces challenges due to large intra-class differences, small inter-class differences, and the complexity of fine-grained images, which existing methods fail to adequately address, leading to difficulties in feature extraction and classification.
Innovation Solution
A method involving a feature extraction network based on a convolutional neural network, where depth features are divided into groups, with random attention mechanisms to enhance feature saliency and diversity, and a model optimization loss function is calculated to improve discriminative and diverse feature learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image recognition methods are used, then the recognition process is simple, but the accuracy of fine-grained image recognition is low due to large intra-class differences and small inter-class differences
Solution Approach 1:
The patent divides depth features into multiple groups and processes each group separately through random attention mechanisms. This segmentation of feature processing allows the model to capture diverse discriminative features while managing complexity through modular group-wise operations
Solution Approach 2:
The patent introduces a new dimension of feature analysis by creating normalized salient features through softmax normalization and maximum value extraction. This dimensional transformation enables the model to capture feature diversity that complements the discriminative information, addressing both intra-class and inter-class differences
2Measurement precision
If more discriminative features are extracted to reduce intra-class differences, then recognition accuracy improves, but the model learning difficulty increases due to small inter-class differences
Solution Approach 1:
By segmenting depth features into groups and applying random attention mechanisms, the model learns discriminative features in a distributed manner across groups. This reduces learning difficulty compared to processing all features uniformly, while still achieving high feature discrimination through the aggregation of group-wise salient features
Solution Approach 2:
The patent transforms the feature representation by applying softmax normalization and extracting maximum values to create normalized salient features. This parameter transformation enhances feature diversity and makes the learning task more tractable by emphasizing the most salient characteristics in each feature group
3Adaptability or versatility
If traditional feature extraction methods are used, then the processing speed is fast, but the ability to capture diverse features for fine-grained classification is insufficient
Solution Approach 1:
The patent segments feature processing into group-wise operations with random attention mechanisms, which captures feature diversity efficiently. The segmentation allows parallel processing of multiple feature groups, maintaining processing efficiency while enhancing feature diversity through the aggregation of diverse group representations
Solution Approach 2:
The patent introduces dynamic elements through random attention mechanisms that adaptively select and emphasize different feature groups during training. This dynamic approach enables the model to capture diverse features adaptively without requiring exhaustive processing of all possible feature combinations, balancing diversity and efficiency
Data Source
AI summary
The present disclosure provides a fine-grained image recognition method, an electronic device and a computer readable storage medium. The method comprises the steps of feature extraction, calculation of feature discriminant loss function, calculation of feature diversity loss function and calculation of model optimization loss function. The present disclosure comprehensively considers influences of factors such as a large intra-class difference, a small inter-class difference, and a great influence of background noise of the fine-grained image, and makes constrains such that the feature maps belonging to each class are discriminative and have the features of corresponding class, thus reducing the intra-class difference, decreasing the learning difficulty and learning better discriminative features. The constraints make the feature maps belonging to each class have a diversity, which increases the inter-class difference, achieves a good result, and is easy for practical deployment, thereby obviously improving the effect of multiple fine-grained image classification tasks.