Image Classification via Display Parameter Transformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image classification methods based on convolutional neural networks require extensive training data and complex network architectures, leading to instability in classification accuracy, especially for specific fields or categories with limited data.
Innovation Solution
The method involves adjusting and transforming image display parameters to create additional training data, constructing a training set with adjusted and transformed images, and training a neural network model to improve classification accuracy and stability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If more training data is collected and network depth is increased to improve classification accuracy, then classification accuracy is improved, but training complexity and device complexity increase
Solution Approach 1:
The patent applies preliminary action by pre-processing original images through adjustment and transformation operations before training. Display parameters of original images are adjusted to satisfy value conditions, and transformations are applied according to distribution conditions to generate diverse training images in advance, eliminating the need for complex network architectures and extensive data collection during training
Solution Approach 2:
The patent applies parameter changes by modifying display parameters of images (such as brightness, contrast, color) to generate varied training samples from limited original images. This transforms a small set of original images into a large diverse training set through parameter manipulation, avoiding the need to increase network depth or collect more data
2Measurement precision
If more training data is collected to improve classification accuracy, then classification accuracy is improved, but loss of time and productivity decrease
Solution Approach 1:
The patent applies copying by generating multiple transformed copies from each original image through display parameter transformations. Instead of collecting numerous original images, the system creates synthetic training samples by copying and transforming existing images, dramatically reducing data collection time while providing sufficient training data
Solution Approach 2:
The patent performs preliminary transformation of images into diverse training samples before the training process. By pre-generating transformed images from original images, the system eliminates time-consuming data collection and preparation during training, accelerating the overall process
3Measurement precision
If network scale is expanded to improve classification accuracy, then classification accuracy is improved, but device complexity and ease of manufacture worsen
Solution Approach 1:
The patent applies parameter changes by modifying image display parameters to generate diverse training samples, which improves classification accuracy without requiring larger network scales. The enhanced variety in training data compensates for smaller network architecture, making model construction simpler and more manufacturable
Data Source
AI summary
Embodiments of the present disclosure provide an image classification method for a computer device. The method includes obtaining an original image and a category of an object included in the original image; adjusting a display parameter of the original image to satisfy a value condition to obtain an adjusted original image; and transforming the display parameter of the original image according to a distribution condition that distribution of the display parameter needs to satisfy, to obtain a transformed image. The method also includes training a neural network model based on the category of the object and a training set constructed by the adjusted original image and the transformed image; and determining a category of an object included in a to-be-predicted image based on the trained neural network model.