Cascaded Neural Networks for Imbalanced Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural networks face challenges in improving the accuracy and efficiency of image processing for detecting abnormalities, such as cancer, in biological organs, particularly due to imbalanced training datasets and the need for enhanced classification capabilities.
Innovation Solution
A method involving training a first neural network using a training dataset, selecting test data that outputs within a specific range, and using this data to train a second neural network, which is optimized for improved classification by adjusting thresholds based on average and standard deviation of output data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single neural network is used for classification, then the device complexity is low, but the classification accuracy and ability to handle imbalanced datasets is insufficient
Solution Approach 1:
The patent divides the classification task into two separate neural networks: a first neural network trained on imbalanced datasets for initial classification, and a second neural network trained on balanced datasets for refined classification. This segmentation allows each network to specialize in specific aspects of the classification problem, improving overall accuracy while managing complexity through functional division.
Solution Approach 2:
The patent implements a cascaded architecture where the output of the first neural network feeds into the second neural network. The second network processes and refines the classifications made by the first network, creating a nested processing structure where each layer builds upon the previous one to achieve higher classification precision.
2Measurement precision
If comprehensive training data is used, then the classification capability is improved, but the computational resource consumption increases
Solution Approach 1:
The training process is segmented into two phases with different data requirements. The first network is trained on imbalanced datasets that require less computational resources, while the second network is trained on selectively balanced data that provides refinement without requiring exhaustive comprehensive training data, thus reducing overall computational burden.
Solution Approach 2:
The first neural network performs preliminary classification on imbalanced datasets before the second network processes the data. This preliminary action filters and pre-processes the data, allowing the second network to focus on refinement tasks with reduced computational requirements compared to training a single network on all data from scratch.
Data Source
AI summary
A method includes: training a first neural network using a first training dataset; inputting each test data of a first test dataset to the first neural network; calculating output data of the first neural network for each test data of the first test dataset; composing a second training dataset of training data from the first test dataset that causes the first neural network to output data within a first range; and training a second neural network using the second training dataset.