Electronic Device Object Recognition Using Selective External Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural network models in electronic devices are limited by CPU and memory capacity, preventing them from classifying all objects and necessitating updates to better suit the device's environment.
Innovation Solution
An electronic device identifies learning images based on low probability values, clusters feature values, and transmits this data to an external device for updating the neural network model using a more accurate model trained on this data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a neural network model is trained to classify only objects designated by the manufacturer in advance, then the model can operate within the device's CPU and memory capacity limits, but it cannot classify all objects and requires updates to better suit the device's environment
Solution Approach 1:
The system pre-identifies learning images with low probability values (below threshold) that represent objects the model struggles to classify. These images are selected in advance for transmission to an external device, allowing the model to be proactively updated with relevant environmental objects before deployment, improving adaptability without compromising reliability
Solution Approach 2:
An external device acts as an intermediary between the electronic device and the neural network model training process. The external device receives learning images, trains or updates the neural network model, and provides updated model information back to the electronic device. This intermediary approach enables model updates without requiring the electronic device to have sufficient computational resources for full model training, resolving the contradiction between adaptability and reliability
2Adaptability or versatility
If learning images are transmitted to an external device for model updating, then the neural network model can be improved with environmental data, but transmission time and communication resources are consumed
Solution Approach 1:
Instead of transmitting all captured images for model updating, the system selectively transmits only those images with low probability values (below a predetermined threshold) that represent problematic classifications. This partial action approach transmits a smaller, more relevant subset of data, reducing transmission time and communication resources while still achieving effective environmental adaptation
Solution Approach 2:
The system changes the parameter of image selection from random or comprehensive sampling to probability-value-based filtering. By using the probability output from the neural network as a selection criterion, the system efficiently identifies and transmits only the most valuable learning images, optimizing the update cycle time while maintaining adaptability
Data Source
AI summary
An electronic device is provide, the electronic device including: a communication interface including at least one circuit; a memory including at least one instruction; and a processor. The processor is configured to: obtain a plurality of images, wherein the plurality of images include an one or more objects; obtain, by inputting the plurality of photographed images into a first neural network model for identifying objects: a feature value for each object of the one or more objects, a predicted class for each object of the one or more objects based on the respective obtained feature values, and a probability value for the predicted class for each of the one or more objects; identify an one or more learning images among the plurality of images based on the obtained probability values; identify one or more clusters of feature values by mapping the feature values of the one or more objects included in the one or more identified learning images to a vector space; obtain a learning data from the one or more identified learning images based on the obtained feature values; transmit the obtained learning data to an external device through the communication interface; receive an information on a second neural network model from the external device, and update the first neural network model based on the received information on the second neural network model.