Neural Network Training Accuracy via Fuzzy Probability Loss
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image detection models using Convolutional Neural Networks (CNNs) face accuracy issues due to inaccurate binary labels during training, leading to low detection performance.
Innovation Solution
The method involves iteratively training multiple neural network models until convergence, using fuzzy probability values and loss parameters to select target samples, and updating the models based on these selections, thereby improving training accuracy through multi-model collaboration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple neural network models are trained iteratively with fuzzy probability values and loss parameters, then image detection accuracy is improved, but device complexity and training time increase
Solution Approach 1:
The patent segments the training process into multiple iterative stages, where in each iteration only certain neural network models are selected for training based on loss parameter thresholds. This divides the complex multi-model training into manageable segments, reducing overall training complexity while maintaining accuracy improvements.
Solution Approach 2:
The patent applies partial action by selectively training only a subset of neural network models in each iteration rather than training all models simultaneously. Models are selected based on whether their loss parameters exceed predefined thresholds, allowing the system to achieve accuracy improvements without the full computational burden of training all models in every iteration.
2Measurement precision
If multiple neural network models are trained iteratively with fuzzy probability values and loss parameters, then image detection accuracy is improved, but training time increases
Solution Approach 1:
The patent implements periodic action through iterative training cycles where models are selected and trained based on periodic evaluation of loss parameters. In each iteration, the system periodically assesses which models exceed threshold values and trains only those models, creating a rhythmic training pattern that reduces total training time compared to continuous training of all models.
Solution Approach 2:
By training only a partial subset of models in each iteration based on loss parameter thresholds, the system reduces the time investment required per iteration while still achieving cumulative accuracy improvements across multiple iterations.
3Ease of manufacture
If simple binary labels are used for training samples, then training process is simplified, but detection accuracy deteriorates due to inaccurate labels
Solution Approach 1:
The patent changes the parameter representation from simple binary labels to fuzzy probability values that capture uncertainty in labeling. Instead of using discrete 0 or 1 labels, the system employs continuous probability values between 0 and 1, allowing the training process to account for label ambiguity while maintaining relative simplicity in the training framework.
Solution Approach 2:
The patent introduces loss parameters as an intermediary mechanism that bridges the gap between simple binary labels and accurate detection. The loss parameters serve as mediators that quantify the discrepancy between predicted fuzzy probabilities and target labels, enabling the system to handle label inaccuracies without requiring complex labeled data.
Data Source
AI summary
Disclosed herein are an image detection method and apparatus, a computer-readable storage medium, and a computer device. The method includes iteratively training a plurality of neural network models to obtain a plurality of trained neural network model; and performing detection on an image to be detected using the trained plurality of neural network models to obtain a detection result. Each iteration of training includes: for each of a plurality of sample images, separately inputting the sample image into the neural network models to obtain a fuzzy probability value set, and calculating, based on the fuzzy probability value set and preset label information of the sample image, a loss parameter of the sample image; selecting target sample images based on a distribution of loss parameters of the plurality of sample images; and updating the plurality of neural network models based on the target sample images.