Neural Network Object Classification Using Reduced Pixel Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial neural network (ANN) algorithms struggle to efficiently classify pixel data, especially in applications with limited computational resources and high-resolution images, where they fail to maintain effectiveness.
Innovation Solution
A computer-implemented method using two separate neural networks to classify pixel data. The first neural network determines the center and stride for a kernel with predetermined size and coefficients, filtering the pixel data to obtain reduced pixel data with a fixed resolution. The second neural network then classifies this reduced data, focusing only on the region of interest.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a standard artificial neural network is used to classify high-resolution pixel data, then classification accuracy is maintained, but computational resource consumption increases and processing speed decreases
Solution Approach 1:
The patent applies segmentation by dividing the high-resolution pixel data into multiple patches or regions of interest. Instead of processing the entire high-resolution image at once, the system segments it into smaller manageable parts that can be processed efficiently by the neural network, thereby maintaining accuracy while reducing computational burden and improving processing speed.
Solution Approach 2:
The patent extracts only the relevant regions of interest from the high-resolution pixel data using object proposal algorithms. By taking out and focusing only on the significant portions of the image rather than processing all pixels, the system achieves accurate classification with reduced computational resource consumption and faster processing.
2Measurement precision
If the pixel data resolution is increased, then classification accuracy is improved, but memory requirements and computational complexity increase
Solution Approach 1:
The patent segments high-resolution pixel data into smaller patches, allowing the system to maintain access to high-resolution information where needed while reducing the overall computational complexity. Each patch can be processed independently with lower memory requirements, yet the collective analysis maintains high classification accuracy.
Solution Approach 2:
The patent applies partial action by processing only the necessary portions of the high-resolution image (regions of interest) rather than the entire dataset. This selective processing approach maintains classification accuracy by focusing computational resources on relevant areas while reducing overall computational complexity and memory requirements.
3Reliability
If the entire pixel data is processed for classification, then complete information is analyzed, but processing time and resource consumption increase
Solution Approach 1:
The patent extracts and processes only the regions of interest from the complete pixel data using object proposal algorithms. By taking out the relevant portions that contain actual objects or features of interest, the system maintains classification completeness for meaningful content while dramatically reducing processing time and resource consumption by ignoring irrelevant background areas.
Solution Approach 2:
The patent applies partial action by processing only the necessary regions of interest rather than the entire pixel data. This approach ensures that all relevant information for classification is analyzed while avoiding waste of computational resources on unnecessary areas, thereby maintaining reliability while reducing processing time.
Data Source
AI summary
A computer-implemented method for classifying pixel data comprising an object of an object type. The method includes determining, by a first neural network, from the pixel data, a center and a stride for a predetermined kernel, wherein the kernel comprises at least part of the object when placed over the pixel data in accordance with the center and the stride; filtering the pixel data by subsequently convolving the pixel data with the kernel at locations defined by the center and stride, to obtain reduced pixel data thereby comprising at least part of the object, wherein the reduced pixel data has a predetermined resolution smaller than a data resolution of the pixel data; and classifying, by a second neural network, the reduced pixel data to the object type, thereby obtaining an object type classification.