Spot Defect Recognition via Positional Encoding and Attention Layers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for recognizing small targets, such as spot defects in industrial production, face challenges in ensuring high recognition rates and accuracy due to factors like small target sizes and complexity.
Innovation Solution
A recognition method utilizing a deep learning-based model with a self-attention mechanism, including feature determination layers and positional encoding, to enhance the detection of spot defects by generating attention features and determining positional and classification information through a series of neural network layers and attention calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional recognition methods are used for small targets, then the device complexity is low, but the recognition accuracy and recognition rate deteriorate
Solution Approach 1:
The recognition model is divided into multiple feature determination layers (first feature determination layer, second feature determination layer, spot defect determination layer) with each layer performing specific feature extraction and processing tasks. This segmentation allows the complex recognition task to be broken down into manageable stages, improving accuracy while organizing complexity systematically
Solution Approach 2:
Positional encoding is introduced to add positional information dimension to the feature maps. The positional encoding layer transforms spatial position information into a format that can be integrated with feature data, enabling the model to understand target locations without increasing base model complexity
2Reliability
If traditional recognition methods are used for small targets, then the processing speed is fast, but the recognition rate deteriorates
Solution Approach 1:
The first feature determination layer performs preliminary feature extraction and the second feature determination layer performs further feature processing before the final spot defect determination. This preliminary action prepares the data in advance, allowing the final recognition stage to operate more efficiently with pre-processed features
Solution Approach 2:
The positional encoding layer acts as an intermediary between the feature extraction layers and the spot defect determination layer. It transforms and integrates positional information into the feature maps, mediating the information flow to improve both accuracy and processing efficiency
3Measurement precision
If small target recognition is performed without positional encoding, then the model complexity is low, but the ability to recognize target position deteriorates
Solution Approach 1:
The positional encoding layer serves as an intermediary component that specifically handles positional information. It transforms coordinate data into encoded features that can be seamlessly integrated with the main feature extraction pipeline, adding position recognition capability without requiring complete model redesign
Solution Approach 2:
Positional encoding applies localized transformations to specific position coordinates rather than processing the entire image globally. Each position in the feature map receives encoding based on its local coordinates, enabling precise position recognition while keeping the computational complexity localized rather than global
Data Source
AI summary
The present disclosure provides a method for recognizing a small target based on a deep learning network. The method comprises: determining, based on a collected image, spot defect information through a recognition model including a first feature determination layer, a second feature determination layer, and a spot defect determination layer, determining, based on the collected image, a first feature map, determining, based on the first feature map, a second feature map by fusing with positional encoding, determining, based on the second feature map, a third feature map through the second feature determination layer, and obtaining, based on the third feature map, positional information of the spot defect through a first determination layer, and determining, based on the third feature map, classification information of the spot defect through a second determination layer.