Stacked Feature Extraction Element for On-Sensor CNN Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Feature extraction processing in image analysis is time-consuming due to the need for data conversion and transfer, limiting the acceleration of this process.
Innovation Solution
A feature extracting element with a light-receiving substrate and laminated substrates containing convolution and pooling processing units, along with a controlling unit that repeats convolution operations using predetermined filter coefficients, allowing direct processing of pixel values and reducing processing time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If feature extraction is performed by dividing an image into blocks and processing per block, then processing time is reduced, but the overall acceleration is insufficient due to image generation and data transfer overhead
Solution Approach 1:
The patent merges the image sensor and feature extraction processing unit into a single integrated device. The light-receiving substrate captures images while the laminated substrates perform feature extraction processing directly on the captured data, eliminating separate image generation and data transfer steps. This integration combines previously separate functions into one unified system, resolving the time loss from data transfer between separate components.
Solution Approach 2:
The patent introduces an intermediate processing layer between image capture and full image processing. The feature extraction element performs preliminary processing on blocks of pixel data before complete image generation, acting as an intermediary that reduces the data volume and processing complexity for subsequent stages. This intermediary processing accelerates feature extraction by working on smaller data units before full image assembly.
2Measurement precision
If more pixels are added to increase resolution, then image quality improves, but processing time increases due to larger data volume
Solution Approach 1:
The patent segments the image data into blocks and processes each block independently through the laminated substrates. Instead of processing the entire high-resolution image at once, the system divides the pixel array into manageable blocks that can be processed in parallel through the convolution and pooling layers. This segmentation maintains high resolution capability while reducing the time complexity of processing by working on smaller units simultaneously.
Solution Approach 2:
The patent adds a spatial dimension to processing by using multiple laminated substrates arranged in layers. The convolution processing unit and pooling processing unit are distributed across different substrate layers, creating a three-dimensional processing architecture. This dimensional transformation allows parallel processing operations to occur simultaneously across layers, reducing overall processing time while maintaining high pixel count capability.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach significantly reduces processing time and miniaturizes devices by eliminating data storage and transfer processes, while maintaining high processing speed even with increased pixel numbers, enabling efficient feature extraction for image analysis.
Implementation Method 1
a light-receiving substrate where a plurality of light-receiving elements for photoelectrically converting received light are two-dimensionally arrayed
Data Source
AI summary
An element includes a plurality of light-receiving elements to photoelectrically convert light received from an object, a convolution processing unit to perform convolution operation on signals that are output from the plurality of light-receiving elements, and a pooling processing unit to sample a signal that is output from the convolution processing unit, based on a predetermined condition. The convolution operation of the convolution processing unit and the sampling of the pooling processing unit are repeated.