Imaging Element With Analog CNN Circuits for Real-Time Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolutional neural networks (CNNs) in image recognition systems face bottlenecks in convolution and pooling operations, leading to increased lead times and system complexity, making real-time processing difficult.
Innovation Solution
An imaging element with integrated convolution and pooling circuits that perform these operations in an analog manner within the imaging element, reducing the need for external high-specification GPUs and simplifying the image recognition system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If convolution and pooling operations are performed using conventional digital processing methods, then image recognition accuracy can be maintained, but processing time increases and real-time recognition becomes difficult
Solution Approach 1:
The patent replaces digital mechanical processing (sequential computation) with optical field-based parallel processing. Convolution operations are performed using optical filters and field correlations, enabling simultaneous processing of multiple pixel signals through physical field interactions rather than sequential digital computation, thereby reducing processing time while maintaining recognition accuracy
Solution Approach 2:
The patent performs convolution operations in advance during the image capture phase using optical filters positioned at the pixel array. By pre-processing the optical field before digital conversion, the system prepares feature-enhanced images that require minimal subsequent digital processing, significantly reducing the time required for real-time recognition while preserving accuracy
2Loss of time
If convolution and pooling operations are performed in a realistic time, then lead time for image recognition is reduced, but system complexity increases
Solution Approach 1:
The patent substitutes complex digital signal processing systems with simpler optical field processing mechanisms. By using optical filters, lenses, and field correlations to perform convolution operations physically, the system achieves rapid processing without requiring complex digital circuitry or large-scale computational hardware, thus reducing lead time while keeping the system relatively simple
Solution Approach 2:
The optical field automatically performs convolution and correlation operations through physical interactions between light waves and optical filters. The system leverages the natural properties of optical fields to execute processing tasks without requiring extensive external control or complex algorithmic intervention, reducing both processing time and system complexity
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
The solution allows for faster image recognition processing times and simplifies the image recognition system by performing convolution and pooling operations within the imaging element, reducing memory and logic processing loads.
Implementation Method 1
an imaging section in which a plurality of pixels individually including a photoelectric conversion element is arranged in a matrix
Data Source
AI summary
Provided is an imaging element according to an embodiment of the present disclosure that includes an imaging section in which a plurality of pixels individually including a photoelectric conversion element is arranged in a matrix, a convolution circuit that performs convolution processing on a plurality of pixel signals, which are analog signals each output from the plurality of pixels, on a basis of a convolution coefficient, and a pooling circuit that performs pooling processing on the plurality of pixel signals that has been subjected to the convolution processing.