Image Sensor Pooling Circuit for Faster Low-Power Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Imaging devices with solid-state imaging elements face challenges in processing time and power consumption due to increased arithmetic operations required for advanced image processing, particularly in in-vehicle systems where speed and efficiency are critical.
Innovation Solution
The imaging device incorporates a novel structure with a pooling processing function of a neural network, including a pixel region with a pooling module that performs pooling processing based on the number of pixels, reducing arithmetic operations and power consumption by selectively outputting and processing only the most significant signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If advanced image processing with increased arithmetic operations is performed, then image recognition accuracy is improved, but processing time is increased
Solution Approach 1:
The patent extracts and processes only the most significant signals from image data through pooling operations, rather than processing all pixel data. This selective extraction reduces the arithmetic operations needed while maintaining recognition accuracy, directly resolving the contradiction between accuracy and processing time.
Solution Approach 2:
The patent applies partial action by performing pooling processing that selects only the maximum or average values from regions of interest, rather than processing the entire image data set. This partial processing approach maintains sufficient accuracy for recognition while dramatically reducing processing time and computational load.
2Measurement precision
If advanced image processing with increased arithmetic operations is performed, then image recognition capability is improved, but power consumption is increased
Solution Approach 1:
The patent extracts only the most significant signals through pooling operations before neural network processing, reducing the total arithmetic operations required. This extraction approach maintains recognition capability while reducing power consumption by minimizing the computational workload on energy-consuming components.
Solution Approach 2:
By performing partial processing through pooling operations that select only essential signal features, the patent reduces the overall arithmetic operations needed in the neural network, thereby reducing power consumption while maintaining sufficient recognition capability.
3Loss of information
If all pixel data is processed, then complete image information is obtained, but arithmetic amount is increased
Solution Approach 1:
The patent extracts the most significant signals from pixel data through pooling operations, obtaining sufficient image information for recognition without processing all pixel values. This extraction maintains information completeness for recognition purposes while reducing the arithmetic amount.
Solution Approach 2:
The patent applies partial processing by performing pooling operations that select only the necessary signal features (maximum or average values) from pixel regions, obtaining sufficient image information while reducing the arithmetic operations required compared to processing all pixel data.
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 power consumption by minimizing the data processed and transferred to the neural network, enhancing the efficiency of image recognition systems.
Implementation Method 1
the pixel has a function of obtaining a first signal through photoelectric conversion
Data Source
AI summary
An imaging device that facilitates pooling processing. A pixel region includes a plurality of pooling modules and an output circuit, the pooling module includes a pooling circuit and a comparison module, the pooling circuit includes a plurality of pixels and an arithmetic circuit, and the comparison module includes a plurality of comparison circuits and a determination circuit. The pixel can obtain a first signal through photoelectric conversion, and can multiply the first signal by a given scaling factor to generate a second signal. The pooling circuit adds a plurality of second signals in the arithmetic circuit to generate a third signal, the comparison module compares a plurality of third signals and outputs the largest third signal to the determination circuit, and the determination circuit determines the largest third signal and binarizes it to generate a fourth signal. In the imaging device, the pooling module performs pooling processing in accordance with the number of pixels and outputs data obtained by the pooling processing.