Multi-Scale CNN Anomaly Detection for Real-Time Image Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection methods require a relatively long time to achieve high accuracy, which hinders real-time performance.
Innovation Solution
An anomaly detection method utilizing a convolutional neural network that obtains and processes first and second feature data of different sizes through N and M convolutional layers respectively, allowing for efficient anomaly detection by leveraging the strengths of each feature size for accurate and rapid analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single size feature map is used for anomaly detection, then the detection process is simple, but the detection accuracy is insufficient
Solution Approach 1:
The patent segments the feature extraction process by obtaining feature maps at multiple different sizes from the convolutional neural network (first feature data from N layers, second feature data from M layers). This segmentation allows the system to analyze anomalies at different spatial resolutions, improving detection accuracy while managing complexity through structured multi-scale processing
Solution Approach 2:
The patent adds the dimension of spatial scale by processing feature data at multiple sizes simultaneously. By combining feature maps from different convolutional layers (different downsampling levels), the system captures both fine-grained and coarse-grained anomaly characteristics, resolving the contradiction between accuracy and complexity
2Measurement precision
If high-accuracy anomaly detection is performed using detailed feature analysis, then detection precision improves, but processing time increases
Solution Approach 1:
The patent performs preliminary feature extraction at multiple scales during the forward propagation phase of the CNN. By pre-obtaining first and second feature data of different sizes before anomaly detection, the system prepares processed information in advance, enabling rapid comparison and detection without time-consuming computations during the actual detection phase
Solution Approach 2:
The patent creates multiple copies of the input image through different convolutional layer processing, generating feature maps at various sizes. These copied and transformed feature representations allow parallel analysis at different resolutions, achieving high accuracy detection without sequential processing that would increase time consumption
3Measurement precision
If multiple feature sizes are processed simultaneously, then detection accuracy improves, but computational requirements increase
Solution Approach 1:
The patent makes the convolutional neural network multi-functional by using the same network structure to extract features at multiple scales for different analysis purposes. The encoder-decoder architecture processes first and second feature data through shared computational pathways, reducing redundant computations and energy consumption while maintaining multi-scale analysis capabilities
Solution Approach 2:
The patent changes the parameter of feature map size by selectively processing features from different convolutional layers (different N and M values). This parameter variation allows the system to adapt computational resources to the specific detection needs, balancing accuracy requirements with energy consumption by processing only necessary feature scales
4Measurement precision
If feature data from multiple convolutional layers is obtained, then anomaly detection capability improves, but memory requirements increase
Solution Approach 1:
The patent extracts only the essential feature data from intermediate convolutional layers (first feature data from N layers, second feature data from M layers) rather than processing all possible layer outputs. This selective extraction reduces memory consumption by retaining only the most informative feature representations needed for accurate anomaly detection
Solution Approach 2:
The patent implements a nested structure where feature maps from different convolutional layers are hierarchically organized and processed. The encoder-decoder architecture nests multiple feature processing operations, allowing efficient memory management by reusing computational results across different feature scales and reducing redundant storage requirements
Data Source
AI summary
An anomaly detection method by which a computer performs anomaly detection includes: obtaining first feature data outputted through N (N is an integer not less than 1) convolutional layers of a convolutional neural network configured as an encoder when an image is inputted to the convolutional neural network; obtaining second feature data outputted through M (M is an integer not less than 1, and M≠N) convolutional layers of the convolutional neural network and different in size from the first feature data; and performing anomaly detection on the image by using features indicated by the first feature data and the second feature data that are different in size.