Parallel CNN Processing for Large Image Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolutional neural networks face increased processing time and reduced capability to extract small features when handling large images, such as high-resolution or panoramic images, due to the substantial number of calculations required, which is exacerbated by the need to reduce image resolution to manage processing efficiently.
Innovation Solution
The method involves dividing image data into overlapping portions and processing each portion through separate circuitries using a limited number of layers, followed by combining these outputs and further processing the combined output through additional layers, allowing for parallel processing and reduced total processing time while maintaining the ability to extract small features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If the resolution of the image is reduced to decrease the number of calculations, then the processing time is reduced, but the possibility to extract small features is reduced
Solution Approach 1:
The image is divided into multiple portions that are processed in parallel by separate circuitries. Each circuitry processes a portion through a first number of layers independently, enabling parallel processing that reduces overall processing time while maintaining the ability to detect small features in each segment.
Solution Approach 2:
The patent introduces a spatial dimension by processing multiple image portions simultaneously across different circuitries rather than sequentially through all layers. This parallel processing architecture effectively adds a dimensional aspect to the computation, reducing time complexity while preserving feature detection capability.
2Productivity
If the resolution of the image is reduced to manage processing efficiently, then the processing efficiency is improved, but the capability to extract small features is reduced
Solution Approach 1:
The image processing task is segmented into multiple portions handled by different circuitries. This segmentation allows efficient parallel processing of large images without requiring significant resolution reduction, as each circuitry processes a manageable portion while contributing to the overall feature extraction.
Solution Approach 2:
The outputs from multiple circuitries processing different image portions are combined to form a comprehensive result. This merging approach allows the system to process large images efficiently through parallel computation while maintaining the capability to extract small features by aggregating information from all portions.
3Area of stationary object
If the image size is increased to represent a wider scene, then the scene coverage is improved, but the processing time is substantially increased
Solution Approach 1:
Large images representing wide scenes are divided into multiple portions that can be processed in parallel. This segmentation enables the system to handle extensive scene coverage by distributing the processing load across multiple circuitries, thereby reducing the overall processing time despite the increased image size.
Solution Approach 2:
The patent leverages parallel processing across multiple circuitries to add a computational dimension that reduces time complexity. By processing multiple portions of large images simultaneously rather than sequentially, the system achieves efficient handling of wide-scene images without proportionally increasing processing time.
Data Source
AI summary
A method (100), a device (600;700) and a system (800) for processing image data representing a scene for extracting features related to objects in the scene using a convolutional neural network are disclosed. Two or more portions of the image data representing a respective one of two or more portions of the scene are processed (S110), by means of a respective one of two or more circuitries, through a first number of layers of the convolutional neural network to form two or more outputs, wherein the two or more portions of the scene are partially overlapping. The two or more outputs are combined (S120) to form a combined output, and the combined output is processed (S130) through a second number of layers of the convolutional neural network by means of one of the two or more circuitries for extracting features related to objects in the scene.