AI Image Recognition with Hierarchical Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image recognition technologies face limitations in accuracy and fail to efficiently reflect field data features across various technical fields, necessitating a solution for improved performance.
Innovation Solution
An electronic device equipped with a processor that utilizes a trained artificial intelligence model with multiple classifiers, performing backpropagation to minimize total loss function values and convert images into gray scale for enhanced classification, incorporating convolutional and recurrent neural networks for accurate image recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single classifier is used for image recognition, then the device complexity is low, but the recognition accuracy is limited
Solution Approach 1:
The patent divides the image recognition task into multiple classification layers (first class and second class) with different granularities. The first class classifier performs coarse-grained classification, while the second class classifier performs fine-grained classification. This segmentation allows the system to achieve high accuracy without requiring a single overly complex classifier, as each classifier focuses on a specific classification level.
Solution Approach 2:
The patent introduces a hierarchical dimension to the classification process by organizing classes into multiple levels (first class with broader categories, second class with more specific subcategories). This dimensional organization allows the system to process information at different levels of abstraction, improving overall recognition accuracy while maintaining manageable complexity at each level.
2Measurement precision
If multiple classifiers with multiple classes are used, then the recognition accuracy improves, but the computational time increases
Solution Approach 1:
The patent performs preliminary coarse-grained classification using the first class classifier before proceeding to fine-grained classification with the second class classifier. This preliminary action quickly eliminates many incorrect categories, reducing the search space for subsequent detailed classification and thereby reducing overall computational time while maintaining high accuracy.
Solution Approach 2:
By segmenting the classification process into multiple stages with different levels of detail, the system avoids the computational burden of a single comprehensive classifier. Each stage processes information at an appropriate level of granularity, optimizing the balance between accuracy and computational efficiency.
3Adaptability or versatility
If general image recognition models are used, then the model structure is simple, but field data features are not efficiently reflected
Solution Approach 1:
The patent applies local quality by designing classification systems tailored to specific application domains. The first and second classes are organized according to domain-specific taxonomies and requirements, allowing the model to efficiently reflect and utilize field data features relevant to particular industrial applications while maintaining a relatively simple overall architecture.
Data Source
AI summary
Disclosed are an electronic device and a method for controlling an electronic device. Specifically, the present disclosure relates to: an electronic device configured to input an acquired image into a trained artificial intelligence model, acquire information about the image from a plurality of classifiers which are included in the artificial intelligence model and correspond to a plurality of layers classified according to higher and lower concepts of an object included in the image, train the artificial intelligence model on the basis of the information about the acquired image, and perform image recognition by using the trained artificial intelligence model; and a method for controlling an electronic device.