Multi-Model Image Recognition Switching for Embedded Reliability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image capture devices with embedded convolutional neural networks face challenges in setting optimal image capture conditions, particularly for recognizing small targets in high-resolution images, due to limited computational capacity.
Innovation Solution
A recognition process execution program and device that utilize multiple trained models and reliability determination functions to manage processing loads, adjusting image capture conditions and resolutions based on reliability thresholds to enhance performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If a first recognition process with lower processing load is executed on an embedded device, then the device can operate within its limited computational capacity, but the recognition reliability may be insufficient for small targets in high-resolution images
Solution Approach 1:
The system dynamically switches between the first recognition process (lower processing load) and the second recognition process (higher processing load) based on real-time reliability assessment. When the determination function detects that recognition reliability is insufficient, it automatically transitions to the second process, and when reliability is sufficient, it returns to the first process. This dynamic adaptation allows the embedded device to optimize between power consumption and recognition reliability.
Solution Approach 2:
The determination function serves as a feedback mechanism that continuously evaluates the output of the first recognition process and decides whether to invoke the second recognition process. This feedback loop ensures that the system maintains appropriate recognition reliability while managing computational resources effectively. The feedback mechanism compares recognition results against reliability thresholds and triggers process switching accordingly.
2Reliability
If a second recognition process with higher processing load is executed to improve recognition reliability, then recognition accuracy for small targets improves, but the computational capacity requirements exceed what the embedded device can provide
Solution Approach 1:
The recognition system is segmented into two distinct processes: a first recognition process with lower processing load for normal operation, and a second recognition process with higher processing load for enhanced reliability when needed. This segmentation allows the embedded device to primarily operate with the lighter first process while having the capability to switch to the more demanding second process only when reliability requirements demand it, thus managing computational capacity effectively.
Solution Approach 2:
The system employs dynamic process switching based on real-time reliability assessment. The determination function monitors recognition quality and dynamically selects between the first and second recognition processes. This dynamic approach ensures that the computationally intensive second process is executed only when necessary, rather than continuously, thereby fitting within the embedded device's computational capacity constraints while maintaining high reliability when needed.
3Reliability
If image resolution is increased to improve target recognition, then recognition reliability for small targets improves, but the processing load increases beyond embedded device capabilities
Solution Approach 1:
The system applies different processing qualities to different situations: the first recognition process handles normal-quality image processing for most cases, while the second recognition process is activated only when high-quality processing is needed for small targets. This local quality approach ensures that high-resolution processing is applied selectively rather than universally, managing processing load within embedded device capabilities while maintaining recognition reliability when necessary.
Data Source
AI summary
A recognition process execution program that inputs first image data indicating a first image to a first trained model, makes the first trained model execute a first recognition process for recognizing a target appearing in the first image, makes the first trained model output first reliability data indicating a reliability of a result of the first recognition process, determines a reliability indicated by the first reliability data, and based on a determination result by the determination function, inputs second image data indicating a second image different from the first image data to a second trained model, and makes the second trained model execute a second recognition process, having a processing load greater than or equal to a processing load of the first recognition process, for recognizing the target appearing in the first image and the second image.