Hierarchical Deep CNN for Image Classification via Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep convolutional neural networks (CNNs) face challenges in image classification due to overfitting and inefficiencies in training data, particularly when distinguishing between similar classes, leading to suboptimal classification performance.
Innovation Solution
The implementation of a hierarchical deep CNN (HD-CNN) that employs a coarse-to-fine classification strategy and modular design, utilizing pretraining and fine-tuning with shared shallow layers to improve classification accuracy and reduce overfitting, and a probabilistic averaging layer for combining predictions from multiple branching CNNs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a deep CNN is trained to distinguish between N classes of data, then classification capability is improved, but overfitting occurs and training efficiency deteriorates
Solution Approach 1:
The patent segments the classification task into multiple stages using an ensemble of CNN classifiers. Each classifier is trained on a subset of categories rather than all N classes simultaneously. This segmentation reduces the complexity each individual classifier must handle, preventing overfitting while maintaining overall classification capability across all categories.
Solution Approach 2:
The patent employs partial action by training multiple classifiers on different subsets of the full training set. Each classifier processes only a portion of the data and category combinations, rather than every classifier processing the complete dataset. This approach improves training efficiency while collectively achieving comprehensive classification coverage through the ensemble.
2Reliability
If multiple separate models are used for classification, then classification performance can be improved through averaging, but device complexity increases
Solution Approach 1:
The patent segments the set of N categories into multiple subsets, with each subset handled by a dedicated CNN classifier. This segmentation allows the system to use multiple models for improved performance while keeping each individual model simpler and more manageable. The overall complexity is distributed across multiple specialized classifiers rather than concentrated in one large model.
Solution Approach 2:
Each CNN classifier in the ensemble serves multiple functions: it processes images, performs classification on its assigned subset of categories, and contributes to the overall ensemble prediction. This multi-functionality allows the system to achieve high classification performance through model averaging while maintaining reasonable complexity at the individual classifier level.
Data Source
AI summary
Hierarchical branching deep convolutional neural networks (HD-CNNs) improve existing convolutional neural network (CNN) technology. In a HD-CNN, classes that can be easily distinguished are classified in a higher layer coarse category CNN, while the most difficult classifications are done on lower layer fine category CNNs. Multinomial logistic loss and a novel temporal sparsity penalty may be used in HD-CNN training. The use of multinomial logistic loss and a temporal sparsity penalty causes each branching component to deal with distinct subsets of categories.