Image Recognition Edge Device With Shared Feature Backbone
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning technologies require significant computing resources and time, making it difficult for edge computing platforms to execute machine learning models due to insufficient resources.
Innovation Solution
An image recognition edge device and method that combines a main backbone for feature extraction with a multi-task fully connected layer, allowing multiple models to be combined into one, reducing computing resource consumption while maintaining recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning models are executed on edge computing platforms, then recognition accuracy can be maintained, but computing resources are insufficient
Solution Approach 1:
The patent segments the machine learning model into two distinct parts: a backbone network for feature extraction and multiple task-specific classification heads. This segmentation allows the computationally intensive feature extraction to be performed once, while multiple recognition tasks share the extracted features, thereby reducing overall computing resource consumption on edge devices while maintaining recognition accuracy.
Solution Approach 2:
The backbone network serves as a universal feature extractor that can be shared across multiple different recognition tasks. Instead of deploying separate models for each task, the single backbone provides features for multiple classification heads, making the system multi-functional and resource-efficient on edge computing platforms.
2Adaptability or versatility
If multiple machine learning models are deployed for different recognition tasks, then comprehensive recognition capability is achieved, but computing resource consumption increases
Solution Approach 1:
The patent merges multiple recognition models into a unified architecture where the backbone network is shared across all tasks. The classification heads for different tasks are combined and connected to the common feature extractor, enabling comprehensive recognition capability while avoiding redundant computation that would occur with separate models.
Solution Approach 2:
The unified model architecture provides multi-functionality by enabling multiple recognition tasks to be performed simultaneously using a single deployed model, reducing the overall computing resource consumption compared to deploying multiple separate models on edge devices.
3Measurement precision
If full-resolution images are processed, then recognition accuracy is improved, but computing time and resource consumption increase
Solution Approach 1:
The processing pipeline is segmented into feature extraction and task-specific classification stages. The backbone network processes images to extract features, and then multiple classification heads operate on these extracted features simultaneously. This segmentation enables efficient processing by separating the computationally intensive feature extraction from the lighter classification operations.
Data Source
AI summary
An image recognition edge device is provided, which includes a memory and a processor. The processor accesses multiple instructions to perform the following operations: downsampling an input image to generate a downsampled image; inputting the downsampled image into an object recognition model through multiple convolutional layers thereby sequentially generating multiple feature tensors; selecting a part of the feature tensors to form a first feature tensor pyramid; selecting another part of the feature tensors to form a second feature tensor pyramid; inputting the first feature tensor pyramid and the second feature tensor pyramid into a combined fully connected layer, generating a first image detection label by a first fully connected output layer based on the first feature tensor pyramid, and generating a second image detection label by the second fully connected output layer based on the second feature tensor pyramid.