Image-Quality Adaptive CNN Target Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing target detection methods, both intrusive and non-intrusive, are vulnerable to fake attacks and fail to accurately identify true targets under varying hardware and environmental conditions, leading to erroneous determinations.
Innovation Solution
A method using a cascaded convolutional neural network (CNN) that determines the quality type of an image and selects an appropriate CNN based on image quality parameters, followed by a thresholding process to accurately classify targets as true or false, enhancing robustness and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single convolutional neural network is used for target detection, then the device complexity is reduced, but the detection accuracy deteriorates under varying image quality conditions
Solution Approach 1:
The system dynamically selects from multiple pre-trained convolutional neural networks based on the quality type of the input image. The quality assessment module evaluates image quality parameters and determines the appropriate CNN to use, allowing the detection system to adapt its complexity and performance characteristics to match the actual image conditions rather than using a fixed single-network approach
Solution Approach 2:
The system changes the parameter of network selection by evaluating image quality parameters (such as resolution, lighting conditions, noise levels) and selecting a CNN from multiple pre-trained networks that is optimized for that specific quality type. This parameter-based selection enables the system to maintain high detection accuracy across varying image quality conditions
2Measurement precision
If multiple convolutional neural networks are prepared for different quality types, then the detection accuracy is improved, but the device complexity increases
Solution Approach 1:
Multiple convolutional neural networks are pre-trained offline for different image quality types (e.g., high quality, medium quality, low quality). During actual detection, the system only needs to evaluate image quality and select from these pre-prepared networks, avoiding the complexity of training multiple networks in real-time while still benefiting from having specialized networks for different conditions
Solution Approach 2:
A quality assessment module acts as an intermediary between the input image and the multiple convolutional neural networks. This mediator evaluates the image quality parameters and determines which pre-trained network is most appropriate, simplifying the system architecture by providing a clear selection mechanism rather than requiring complex integration of multiple networks
3Ease of operation
If an artificial design feature extraction method is used, then the ease of operation is improved, but the reliability deteriorates in complex environmental conditions
Solution Approach 1:
The system replaces manual artificial feature design with automated convolutional neural networks that learn features directly from data. Instead of relying on hand-crafted features that may not generalize well to complex conditions, the CNNs automatically learn robust feature representations from training data, improving reliability while maintaining ease of operation through automated processing
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Provided are a method and system for detecting a target, in particular determining if a target is a true target, wherein the method and system take into account a quality of an image. A method of detecting a target includes determining a quality type of an image; determining a convolutional neural network of a quality type corresponding to the quality type of the image; determining a detection value of the image based on the convolutional neural network; and determining whether a target in the image is a true target based on the detection value of the image.