AI Edge Extraction for Blurry Industrial Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing edge extraction algorithms, such as the Canny edge detection algorithm, struggle to accurately detect the real edges of blurry industrial images due to low edge contrast, making them unsuitable for precise industrial applications.
Innovation Solution
A complex industrial image edge extraction method utilizing artificial intelligence, involving image pyramid processing, cropping, scaling, and a trained model comprising convolution modules, BottleneckCSP network modules, spatial pyramid pooling-fast (SPPF) modules, up-sampling modules, and Concat modules, along with preprocessing steps like Gaussian filtering and color space conversion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional gradient-based edge extraction algorithms (e.g., Canny) are used, then the processing speed is fast, but the edge detection accuracy deteriorates for blurry images with low edge contrast
Solution Approach 1:
The patent applies parameter changes by transforming the image data through multiple processing stages: Gaussian smoothing with varying kernel sizes, color space conversions (RGB to HSV), and image pyramid construction with different resolution levels. These parameter transformations enable the AI model to process blurry images effectively by adjusting the representation of edge information at different scales and color spaces, thereby improving edge detection accuracy without excessive complexity
Solution Approach 2:
The patent replaces traditional mechanical gradient-based edge detection algorithms with an artificial intelligence-based system. The AI model (comprising convolution modules, BottleneckCSP network modules, SPPF modules, up-sampling modules, and Concat modules) learns edge patterns from training data and can accurately detect edges in blurry images by recognizing contextual relationships, thereby substituting the mechanical gradient calculation approach with an intelligent pattern recognition system
2Measurement precision
If image processing steps (pyramid processing, cropping, scaling) are added to enhance edge detection, then the edge extraction accuracy improves, but the processing time increases
Solution Approach 1:
The patent applies segmentation by dividing the image processing into distinct functional stages: Gaussian smoothing preprocessing, color space conversion, image pyramid construction, cropping, scaling, and AI model processing. Each stage processes specific aspects of the image data independently, allowing for optimized processing parameters at each level. The image pyramid itself segments the image into multiple resolution levels, enabling the AI model to process edge information at different scales efficiently
Solution Approach 2:
The patent applies preliminary action by performing Gaussian smoothing, color space conversion, and image pyramid construction before the actual edge extraction by the AI model. These preliminary processing steps prepare the image data in advance, transforming it into a format that is more suitable for AI processing. The cropping and scaling operations are also performed beforehand to optimize the input size for the AI model, thereby improving processing efficiency and accuracy
Data Source
AI summary
A complex industrial image edge extraction method based on artificial intelligence and a complex industrial image edge extraction device are provided. The complex industrial image edge extraction method based on artificial intelligence includes steps of obtaining first image data to be subjected to edge extraction, and performing image pyramid processing on the first image data to be subjected to the edge extraction to obtain second image data, performing image cropping processing on the second image data to obtain third image data, performing image scaling processing on the third image data to obtain first scaled image data, inputting the first scaled image data into an artificial intelligence model for training and convergence and performing edge extraction processing, and then outputting an edge extraction result of the first scaled image data.