Convolutional Neural Network Stepwise Feature Kernel Application
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolutional neural networks require significant computational resources when applied to images larger than those used during training, as they need to be applied at every possible offset, making them inefficient for handling higher image resolutions.
Innovation Solution
A method where a convolutional neural network is trained with images of a first size and then applied to images of a second size, using a feature kernel that is repeatedly applied in a stepwise manner with a step size greater than half of the original feature map size, reducing the number of calculations needed to cover the entire feature map.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the convolutional neural network is applied at every possible offset to handle images larger than training images, then the feature detection accuracy is improved, but the computational complexity increases significantly
Solution Approach 1:
The patent divides the large image into multiple smaller patches that match the training image size. Each patch is processed independently by the trained convolutional neural network, and the results are aggregated to form the final detection output. This segmentation approach maintains detection accuracy while significantly reducing computational complexity compared to processing the entire large image at once.
2Measurement precision
If the convolutional neural network is applied at every possible offset to handle images larger than training images, then the feature detection accuracy is improved, but the processing time increases significantly
Solution Approach 1:
The patent segments the large image into overlapping or non-overlapping patches, processes each patch independently through the trained network, and aggregates results. This parallelizable approach dramatically reduces processing time compared to applying the network at every possible offset across the entire large image, while maintaining detection accuracy through proper patch aggregation.
3Area of stationary object
If the feature kernel size is increased to cover larger areas, then the detection coverage is improved, but the number of calculations increases
Solution Approach 1:
Instead of using a single large feature kernel that would require numerous calculations, the patent segments the processing into multiple smaller kernels applied to different patches. Each small kernel performs fewer calculations, but the collective coverage across all patches achieves the same or better detection coverage, significantly reducing the total number of calculations required.
Data Source
AI summary
A method and an image processing entity for applying a convolutional neural network to an image are disclosed. The image processing entity processes the image while using the convolutional kernel to render a feature map, whereby a second feature map size of the feature map is greater than a first feature map size of the feature maps with which the feature kernel was trained. Furthermore, the image processing entity repeatedly applies the feature kernel to the feature map in a stepwise manner, wherein the feature kernel was trained to identify the feature based on the feature maps of the first feature maps, wherein the feature kernel has the first feature map size.