Assembly Change Detection Using Attention Mechanism and Depth Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current change detection methods are inadequate for multi-view assembly processes of mechanical assembly bodies due to complex structures, severe occlusion, large viewing angle changes, and limited data sets, leading to low accuracy and efficiency in detecting errors during assembly.
Innovation Solution
An assembly body change detection method based on an attention mechanism, which involves establishing a 3D model, performing imaging processing, training a detection model using semantic segmentation and feature extraction with a RepVGG classification network, and employing an attention mechanism to capture context information and calculate Euclidean distances between feature pictures, outputting change images to identify errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional change detection methods using difference operators are used, then the process is simple, but the accuracy is low due to noise interference in difference image formation
Solution Approach 1:
The patent replaces traditional mechanical difference operators with a deep learning-based detection model that uses semantic segmentation and attention mechanisms. This substitution eliminates the noise interference problem inherent in traditional difference image formation while achieving higher change detection accuracy through learned feature representations.
Solution Approach 2:
The patent introduces semantic segmentation as an intermediary step between input images and change detection. By first segmenting images into semantic regions and then comparing segmented results, the system avoids direct pixel-level difference operations that cause noise, thereby improving accuracy while maintaining manageable complexity.
2Measurement precision
If deep learning methods are used for change detection, then accuracy improves, but the complexity of the detection model increases significantly
Solution Approach 1:
The patent segments the deep learning model into distinct functional modules: semantic segmentation network, attention mechanism module, and change detection module. This segmentation allows each component to be optimized independently and facilitates efficient training and deployment, reducing overall system complexity while maintaining high accuracy.
Solution Approach 2:
The patent performs semantic segmentation as a preliminary action before change detection. By pre-processing images to extract semantic information, the subsequent change detection operates on simplified semantic representations rather than raw pixels, reducing the complexity of the final detection step while improving accuracy.
3Adaptability or versatility
If multi-view images with large viewing angle changes are used, then comprehensive detection coverage is achieved, but the difficulty of detecting changes increases due to severe occlusion and structure variations
Solution Approach 1:
The patent transforms the problem from 2D image space to 3D point cloud space by using depth information. This dimensional change allows the model to reason about spatial relationships and occlusions more effectively, as the 3D representation preserves geometric structures that are lost in 2D projections, thereby reducing detection difficulty across multiple views.
Solution Approach 2:
The patent changes the parameter space from pixel intensities to geometric features and semantic attributes. By transforming images into feature representations that capture essential geometric and semantic properties, the model becomes more robust to viewing angle changes and occlusions, reducing the difficulty of detection across diverse multi-view conditions.
Data Source
AI summary
An assembly change detection method based on attention mechanism, including: establishing a three-dimensional model of an assembly body, adding a tag to each part in the three-dimensional model, setting several assembly nodes, obtaining depth images of the three-dimensional model under each assembly node in different viewing angles, and obtaining a change tag image of a added part at each assembly node; selecting two depth images at front and back moments in different viewing angles as training samples; performing semantic fusion, feature extraction, attention mechanism processing and metric learning sequentially on the training samples, training a detection model, continuously selecting training samples to train the detection model, saving model parameters with optimal similarity during training, completing training; and obtaining depth images of successive assembly nodes during assembling the assembly body, inputting depth images into trained detection model, and outputting change image of added part of the assembly body during assembly.