Image Preprocessing Neural Network for Object Detection

Solution Overview

Problem

Existing object recognition technologies for autonomous driving and surveillance systems face challenges in maintaining high accuracy and efficiency due to heat generation and high power consumption, especially when using deep-learning processors.

Innovation Solution

The proposed solution involves an apparatus and method for recognizing objects in images by performing image enhancement processing to increase recognition rates, and then inputting the preprocessed images to an artificial neural network for object recognition. This includes training the neural network using feedback results to optimize image preprocessing for improved recognition accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If deep-learning processors are used for object recognition, then recognition accuracy is improved, but power consumption and heat generation increase

Engineering Contradiction:

Improveobject recognition accuracyVSAvoidpower consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system divides the object recognition task into two separate processing stages: a first neural network performs image enhancement and preprocessing, while a second neural network performs the actual object recognition. This segmentation allows each network to be optimized for its specific function, enabling the use of smaller, more energy-efficient networks while maintaining high recognition accuracy through the combined effect of enhanced preprocessing and specialized recognition processing.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The first artificial neural network performs preliminary image enhancement processing before the main object recognition task. This preliminary action of enhancing image quality (through operations like denoising, deblurring, super-resolution) prepares the input data in advance, allowing the second recognition network to operate more efficiently with pre-optimized input, thereby reducing its computational burden and power consumption while improving recognition accuracy.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If deep-learning processors are used for object recognition, then recognition accuracy is improved, but heat generation increases

Engineering Contradiction:

Improveobject recognition accuracyVSAvoidheat generation

Core Design Contradiction:

Measurement precisionVSTemperature

Solution Approach 1:

The system divides the object recognition task into two separate processing stages: a first neural network performs image enhancement and preprocessing, while a second neural network performs the actual object recognition. This segmentation allows each network to be optimized for its specific function, enabling the use of smaller, more energy-efficient networks while maintaining high recognition accuracy through the combined effect of enhanced preprocessing and specialized recognition processing.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The first artificial neural network performs preliminary image enhancement processing before the main object recognition task. This preliminary action of enhancing image quality (through operations like denoising, deblurring, super-resolution) prepares the input data in advance, allowing the second recognition network to operate more efficiently with pre-optimized input, thereby reducing its computational burden and power consumption while improving recognition accuracy.

Inventive Principle:

Principle #10Preliminary action

3Measurement precision

If image enhancement processing is performed, then object recognition rate is improved, but processing complexity increases

Engineering Contradiction:

Improveobject recognition rateVSAvoidprocessing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system divides the object recognition task into two separate processing stages: a first neural network performs image enhancement and preprocessing, while a second neural network performs the actual object recognition. This segmentation allows each network to be optimized for its specific function, enabling the use of smaller, more energy-efficient networks while maintaining high recognition accuracy through the combined effect of enhanced preprocessing and specialized recognition processing.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The first artificial neural network acts as an intermediary between the raw image input and the second recognition network. It performs image enhancement processing that optimizes the input data for the recognition task, serving as a bridge that transforms raw images into enhanced images that are more suitable for accurate object recognition, thereby improving overall system performance while managing complexity through functional separation.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS12327404B2Object detection apparatus using an image preprocessing artificial neural network model

Publication Date: 2025.06.10 DEEPX CO LTD

AI summary

An apparatus for recognizing an object in an image includes a preprocessing module configured to receive an image including an object and to output a preprocessed image by performing image enhancement processing on the received image to improve a recognition rate of the object included in the received image; and an object recognition module configured to recognize the object included in the image by inputting the preprocessed image to an input layer of an artificial neural network for object recognition.