Hierarchical Constraint Loss for Graptolite Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Fine-grained image classification of graptolite specimens is challenging due to subtle differences and complex genetic and evolutionary relationships, leading to poor classification results with existing CNN methods that focus on detailed features, especially in fossil images with unclear textures and false colors.
Innovation Solution
The implementation of Hierarchical Constraint Loss (HC-Loss) in CNN models, which considers the genetic relationships between species to regulate the similarity between image features, reducing the number of parameters and preventing overfitting by weighting the loss function based on the similarity between image pairs, thereby improving classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If CNN uses a large number of intermediate parameters to extract detailed features, then feature representation ability is improved, but the model learns false colors and irrelevant features leading to poor classification results
Solution Approach 1:
The patent changes the parameter space by introducing hierarchical constraint parameters (L2 regularization terms) that control the relationship between intermediate features across different layers. This constrains the model to learn hierarchical feature relationships rather than relying solely on increasing the number of parameters, thereby improving classification accuracy without proportionally increasing model complexity.
Solution Approach 2:
The patent segments the feature extraction process into hierarchical levels with different constraint mechanisms. Intermediate features at different hierarchical levels are subjected to different L2 regularization constraints, allowing the model to learn discriminative features at each level while preventing overfitting to irrelevant details like false colors.
2Measurement precision
If CNN focuses on detailed features to capture subtle differences, then fine-grained discrimination is improved, but the model overfits to false colors and irrelevant details
Solution Approach 1:
The patent introduces L2 regularization parameters that control the magnitude of intermediate features at different hierarchical levels. By adjusting these parameters, the model is guided to focus on reliable discriminative features while suppressing sensitivity to unreliable details such as false colors and lighting variations, thus improving classification robustness.
Solution Approach 2:
The patent implements a feedback mechanism through the loss function that includes L2 regularization terms. The regularization terms provide continuous feedback during training to constrain intermediate features, preventing the model from overfitting to irrelevant details while maintaining its ability to capture subtle discriminative differences.
3Measurement precision
If the model learns complex interactive features to improve classification, then discrimination accuracy is improved, but the number of parameters increases causing overfitting
Solution Approach 1:
The patent changes the parameter optimization strategy by adding L2 regularization terms to the loss function. This modifies the parameter update rule to include a penalty term that prevents parameters from growing excessively large, thereby enabling the model to learn complex hierarchical feature relationships without requiring a proportional increase in the number of parameters.
Data Source
AI summary
The present disclosure relates to a hierarchical constraint (HC) based method and system for classifying fine-grained graptolite images. The method includes: constructing a graptolite fossil dataset; extracting features in graptolite images; calculating the similarity between graptolite images, and performing weighting according to a genetic relationship among species to obtain a weighted HC loss function (HC-Loss) of all graptolite images; calculating cross-entropy loss; taking a weighted sum of HC-Loss and CE-Loss as a total loss function in a training stage; and performing model training. The system of the present disclosure includes a processor and a memory.