Two-Stage Neural Network Classifier for Image Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional image recognition techniques require increased neural network size and calculation, leading to reduced processing speed and accuracy due to the connection of multiple detailed class classifiers to a coarse class classifier, and errors in coarse class classification can result in incorrect detailed classification.

Innovation Solution

A method involving a two-stage learning process where a coarse class classifier and a detailed class classifier are trained separately using neural networks with adjusted final layers to maintain network size, allowing for high-accuracy image recognition without reducing processing speed, by learning features common to coarse and detailed classes respectively.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If multiple detailed class classifiers are connected to a coarse class classifier, then classification accuracy is improved, but processing speed is reduced

Engineering Contradiction:
Improveclassification accuracyVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSSpeed

Solution Approach 1:

The patent segments the classification task into two independent stages: coarse class classification and detailed class classification. Each stage has its own dedicated neural network classifier, avoiding the need to connect multiple detailed class classifiers to a coarse class classifier. This segmentation maintains classification accuracy while improving processing speed by eliminating redundant computational paths.

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If multiple detailed class classifiers are connected to a coarse class classifier, then classification accuracy is improved, but device complexity is increased

Engineering Contradiction:
Improveclassification accuracyVSAvoidneural network structure complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the classification system into two separate, independent neural network classifiers: a coarse class classifier and a detailed class classifier. This segmentation reduces device complexity by eliminating the complex interconnections required in traditional multi-classifier systems, while maintaining high classification accuracy through the two-stage classification approach.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts the classification function into two distinct components, each handling a specific level of classification granularity. By taking out the detailed class classification from the coarse class classification framework and making it an independent system, the overall device complexity is reduced while preserving classification accuracy.

Inventive Principle:
Principle #2Taking out (Extraction)

3Measurement precision

If neural network size is increased to improve classification accuracy, then measurement precision is improved, but processing speed is reduced

Engineering Contradiction:
Improveclassification accuracyVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSSpeed

Solution Approach 1:

The patent segments the classification task into two independent neural networks of moderate size, each optimized for its specific classification level. This avoids the need for a single large neural network, thereby maintaining classification accuracy while improving processing speed by reducing the computational burden on any single network.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3065090B1Learning method and recording medium background
Publication Date: 2020.11.04 PANASONIC INTELLECTUAL PROPERTY MANAGEMENT CO LTD
  • EP3065090B1 patent drawingFigure 1
  • EP3065090B1 patent drawingFigure 2A~2B
  • EP3065090B1 patent drawingFigure 3A~3B

AI summary

Learning method includes performing a first process in which a coarse class classifier configured with a first neural network is made to classify a plurality of images given as a set of images each attached with a label indicating a detailed class into a plurality of coarse classes including a plurality of detailed classes and is then made to learn a first feature that is a feature common in each of the coarse classes, and performing a second process in which a detailed class classifier, configured with a second neural network that is the same in terms of layers other than the final layer as but different in terms of the final layer from the first neural network made to perform the learning in the first process, is made to classify the set of images into detailed classes and learn a second feature of each detailed class.