Neural network model training method and system
By combining convolutional computation and nonlinear convolutional computation in multiple iterations, adjusting the dimensions of image data and using activation functions, the problem of insufficient detection of subtle image features by neural network-like models is solved, and image feature detection with higher sensitivity is achieved.
CN114936634BActive Publication Date: 2026-06-30XINJU TECHNOLOGY CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XINJU TECHNOLOGY CO LTD
- Filing Date
- 2022-04-12
- Publication Date
- 2026-06-30
AI Technical Summary
Technical Problem
Existing neural network-like models lack sensitivity to subtle features in image feature detection, making it difficult to effectively detect images with these subtle features.
Method used
A method combining convolutional computation and nonlinear convolution is adopted. Through multiple iterations of computation and dimensionality adjustment, the results are updated using an activation function, and image feature extraction is performed by combining different mathematical and nonlinear operands.
Benefits of technology
This improves the sensitivity of neural network models to subtle image features, enabling more accurate detection of images with these subtle features, and can be applied to medical imaging and non-destructive testing.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN114936634B_ABST
Abstract
A neural network model training method includes: receiving image data; performing convolution calculation using the image data; performing nonlinear convolution calculation using the image data; and classifying according to the results of the convolution calculation and the nonlinear convolution calculation, wherein the result of the convolution calculation is generated according to a plurality of products of the image data and a mask, and the result of the nonlinear convolution calculation is generated according to the image data, a mask, a mathematical operand and a nonlinear operand. The neural network model training method provided by the present application can train linear and nonlinear operation parameters respectively through deep learning, so that the features extracted by the neural network model have high discrimination, which is beneficial for subsequent classification operation. The neural network model obtained accordingly is more sensitive to some features of the image and can detect images with these subtle features.
Need to check novelty before this filing date? Find Prior Art