Hierarchical Image Decomposition for High-Resolution Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional deep learning methods for object detection struggle with high-resolution images due to resolution mismatches, requiring retraining and increased computational effort, and face challenges like varying light conditions, occlusions, and small defect detection in complex backgrounds.
Innovation Solution
The method decomposes high-resolution images into hierarchically organized layers, applies a pre-trained baseline algorithm to each layer, and performs smart result aggregation through pairwise layer comparison and hierarchical aggregation, allowing for efficient object detection without retraining the algorithm.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional deep learning methods are used for high-resolution image detection, then detection capability is maintained, but computational effort and processing time increase significantly
Solution Approach 1:
The high-resolution image is divided into multiple overlapping patches, which are then processed in parallel. This segmentation approach reduces the computational burden on each individual processing unit while maintaining the overall detection capability through the use of a pyramid structure that aggregates results from different resolution levels.
Solution Approach 2:
The patent introduces a pyramid dimension by creating multiple layers of image patches at different resolutions. Instead of processing a single high-resolution image, the system processes multiple layers where each layer contains patches at a specific resolution, enabling efficient parallel computation while preserving detection accuracy through hierarchical aggregation.
2Measurement precision
If high-resolution images are processed directly, then detection accuracy is maintained, but memory usage and computational resources increase
Solution Approach 1:
The image is segmented into smaller patches that can be processed and stored independently. This reduces the peak memory requirements compared to loading and processing the entire high-resolution image at once, while still maintaining detection accuracy through the overlapping patch design and pyramid aggregation.
Solution Approach 2:
By organizing patches into a pyramid structure with multiple resolution layers, the system distributes memory usage across different layers rather than requiring all high-resolution data to be held in memory simultaneously. This hierarchical organization enables efficient memory management while preserving detection capability.
3Loss of time
If pre-trained baseline algorithms are used without retraining, then development time is reduced, but performance on high-resolution images deteriorates
Solution Approach 1:
The patent creates a pyramid dimension that allows pre-trained algorithms to operate effectively on high-resolution images without retraining. By organizing processed patches into a pyramid structure with multiple resolution layers, the system enables baseline algorithms to leverage their pre-trained knowledge while the hierarchical aggregation recovers fine details from higher resolution layers, maintaining performance without additional training time.
Data Source
AI summary
Aspects of the present invention disclose a method, computer program product, and system for improving object detection in high-resolution images at inference time. The method includes one or more processors receiving a high-resolution image. The method further includes one or more processors decomposing the received image into hierarchically organized layers of images. Each layer comprises at least one image tile of the received image. Each of the image tiles have a corresponding resolution suitable to a baseline image recognition algorithm. The method further includes one or more processors applying the baseline algorithm to each of the image tiles of each layer. The method further includes one or more processors performing a result aggregation of results of the baseline algorithm applications to the image tiles of the layers.