Lightweight Neural Network for Workpiece Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI-based workpiece detection systems require deep network structures, leading to high spatial and temporal costs, making them inefficient for industrial edge devices and requiring frequent retraining for new workpieces with different dimensions.
Innovation Solution
The method involves performing data augmentation on original workpiece images, training a neural network model with multiple feature extraction branches, converting the model into a lightweight detection model, and using this model to detect workpieces with the same shape but different dimensions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep network structures are used for workpiece detection, then detection accuracy is improved, but spatial and temporal costs increase significantly
Solution Approach 1:
The patent divides the original image into multiple image patches and processes them in parallel through the neural network. This segmentation approach allows the deep network to process smaller regions simultaneously, reducing overall detection time while maintaining accuracy through comprehensive coverage of the workpiece.
Solution Approach 2:
The patent performs detection on multiple image patches that may overlap or extend beyond the workpiece boundaries. This partial action approach ensures that the entire workpiece is covered by detection regions, maintaining high detection accuracy while allowing parallel processing to reduce time costs.
2Measurement precision
If deep network structures are used for workpiece detection, then detection accuracy is improved, but resource pressure on edge devices increases
Solution Approach 1:
By segmenting the image into patches and processing them in parallel, the computational load is distributed across multiple smaller tasks. This reduces the peak resource pressure on edge devices while maintaining the accuracy benefits of deep networks through comprehensive workpiece coverage.
3Measurement precision
If model retraining is performed for new workpieces with different dimensions, then detection accuracy is maintained, but production time and costs increase
Solution Approach 1:
The patent creates a universal detection model that can handle workpieces of different dimensions without retraining. The model achieves this by processing images in patches and using data augmentation during training to expose the network to various workpiece sizes, enabling one model to serve multiple detection scenarios.
Solution Approach 2:
The patent uses data augmentation techniques that vary parameters such as workpiece size, position, and orientation in the training data. This teaches the model to be invariant to dimensional changes, allowing it to detect different workpiece variants without retraining while maintaining high accuracy.
Data Source
AI summary
Various embodiments of the teachings herein include a method for detecting a workpiece. An example includes: performing data augmentation on an original image of an original workpiece; performing training on a neural network model comprising multiple feature extraction branches to obtain a workpiece detection model based on a workpiece image group obtained through the data augmentation; converting the workpiece detection model into a lightweight workpiece detection model; and performing detection on workpieces with the same shape and at least one different dimension as the original workpiece based on the lightweight workpiece detection model.