Neural Network Partial Signal Processing for High-Resolution Image Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks face high processing costs and reduced accuracy when handling large images due to excessive reduction in image size in early layers, leading to missed detection of small abnormalities in intruder detection and manufactured product inspection.
Innovation Solution
A processing apparatus that cuts out partial signals from input images, processes them using neural networks with the same layer structure, calculates statistical values for each channel, and generates a synthetic signal by concatenating these values, allowing for high-accuracy processing of large images without reducing image resolution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the image size is excessively decreased in an early layer of the neural network, then the processing amount is reduced, but image resolution is lowered and small abnormality cannot be detected
Solution Approach 1:
The patent divides the large input image into multiple partial images (first partial image, second partial image, etc.) and processes each partial image separately through the neural network. This segmentation allows the network to maintain high resolution for each subset while reducing the overall processing burden compared to processing the entire large image at once.
Solution Approach 2:
The patent introduces a new dimension by processing multiple partial images in parallel through separate neural network paths, then combining their outputs. This dimensional approach allows maintaining detailed information from each partial image while achieving efficient processing through parallel computation.
2Loss of information
If an image having an extremely large size is input to the neural network, then complete image information is preserved, but the processing amount is huge and arithmetic cost is extremely increased
Solution Approach 1:
The patent segments the large input image into multiple smaller partial images that can be processed efficiently by the neural network. This segmentation preserves the complete image information by ensuring each partial image maintains sufficient resolution while reducing the computational burden of processing the entire large image simultaneously.
Solution Approach 2:
The patent merges the processing results from multiple partial images to form the final output. By combining the detected abnormalities or features from each partial image processing path, the system reconstructs comprehensive image information without requiring the neural network to process the entire large image at once, thus reducing arithmetic cost.
Data Source
AI summary
According to an embodiment, a processing apparatus includes a hardware processor. The hardware processor is configured to: cut out, from an input signal, a plurality of partial signals that are predetermined parts in the input signal; execute processing on the plurality of partial signals using neural networks having the same layer structure with each other to generate a plurality of intermediate signals including a plurality of signals corresponding to a plurality of channels; execute predetermined statistical processing on signals for each of the plurality of channels for each of the plurality of intermediate signals corresponding to the plurality of partial signals, to calculate statistics for each channel and generate a concatenated signal by concatenating the statistics of the plurality of respective intermediate signals for each channel; generate a synthetic signal by performing predetermined processing on the concatenated signal; and output an output signal in accordance with the synthetic signal.