Image Classification Using Pseudo-Centroid Datasets for IoT Edge Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
IoT edge devices face challenges in processing and classifying image data due to limited computing resources, requiring significant computing power and memory, which existing technologies have not adequately addressed.
Innovation Solution
An image classification system that uses pseudo-centroid datasets with a reduced number of data points, allowing for efficient storage and processing, where feature values of new image data are compared to these pseudo-centroid data points for classification, along with an image compressing device to reduce computation and storage needs, and reinforcement learning for enhanced accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image data processing is performed through IoT edge device using neural network model, then classification accuracy is improved, but computing resource consumption increases
Solution Approach 1:
The patent segments the image data processing into two parts: (1) feature extraction performed by the IoT edge device using a pre-trained neural network model, and (2) classification performed by comparing extracted features with stored pseudo-centroid datasets. This segmentation allows the computationally intensive training to be done offline while keeping the edge device operations simple and energy-efficient.
Solution Approach 2:
The patent applies preliminary action by pre-training the neural network model offline and generating pseudo-centroid datasets before deployment. The model weights and pseudo-centroids are stored in the edge device, eliminating the need for real-time training and reducing computing resource requirements during actual image classification tasks.
2Measurement precision
If image data processing is performed through IoT edge device using neural network model, then classification accuracy is improved, but memory space requirement increases
Solution Approach 1:
The patent extracts only the essential components needed for classification: pre-trained model weights and pseudo-centroid datasets. By removing unnecessary training data and intermediate processing components, the system achieves accurate classification with minimal memory space requirements suitable for edge devices.
Solution Approach 2:
The patent uses pseudo-centroid datasets as simplified copies of the full training dataset. These pseudo-centroids capture the essential classification information in a compact form, allowing the edge device to perform accurate classification without storing the entire original training dataset.
3Speed
If feature values are compared with pseudo-centroid data points, then processing speed is improved, but data structure complexity increases
Solution Approach 1:
The patent changes the parameter representation from storing complete image data to storing extracted feature vectors and pseudo-centroid representations. This parameter transformation enables faster comparison operations while the data structures remain relatively simple arrays of numerical values, manageable through standard computational techniques.
Data Source
AI summary
An image classification system includes a storage device, a computing device and a first processing device. The storage device stores a plurality of pseudo-centroid datasets, wherein the pseudo-centroid datasets correspond to a plurality of units of first image dataset, and the number of pseudo-centroid data points of each of the pseudo-centroid datasets is much smaller than the number of data points of each of the units of first image dataset. The computing device receives the second image data and computes a plurality of feature values of the second image data. The first processing device receives the feature values and the pseudo-centroid datasets, and compares the feature values with the pseudo-centroid data points to identify and classify the second image data.