Deep CNN Smoke Detection With Bidirectional Multi-Scale Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional smoke detection methods based on low-level features are inaccurate and lack robustness in complex backgrounds, lighting changes, and disturbances, and existing deep convolutional neural networks struggle with feature extraction and multi-scale fusion, leading to unstable detection in complex environments.
Innovation Solution
A disaster smoke detection method using a deep convolutional neural network that performs multiple convolution operations to extract features, enhances them, and applies multi-scale and bidirectional feature fusion to accurately detect smoke and flames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional image processing techniques based on low-level features are used, then the detection method is simple and fast, but the detection accuracy and robustness are greatly reduced in complex backgrounds
Solution Approach 1:
The patent transforms the detection approach from low-level feature parameters (color, shape) to high-level semantic features through deep convolutional neural networks. This parameter transformation enables the system to capture essential smoke characteristics that are invariant to background variations, lighting changes, and disturbances, thereby maintaining high detection accuracy in complex environments while preserving reasonable detection speed through optimized network architecture.
2Measurement precision
If deep convolutional neural networks are used for feature extraction, then the detection accuracy is improved, but the feature extraction and multi-scale fusion capabilities are insufficient in complex environments
Solution Approach 1:
The patent segments the feature extraction process into multiple parallel convolutional pathways, each specialized for detecting smoke at different scales. This segmentation allows the network to simultaneously capture fine-grained details of small smoke particles and broader patterns of large smoke plumes, improving multi-scale feature fusion capability while maintaining detection accuracy in complex environments.
Solution Approach 2:
The patent introduces multi-scale feature maps as an additional dimensional aspect to the standard convolutional network. By processing features at multiple resolution levels and fusing them through specialized fusion layers, the network gains enhanced adaptability to handle smoke targets of varying sizes and distances, thereby improving versatility without sacrificing detection accuracy.
3Measurement precision
If multiple convolution operations and feature fusion are performed, then the feature extraction accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent merges multiple feature maps from different convolutional layers through efficient fusion operations that combine spatial and channel dimensions. This merging strategy consolidates redundant computational operations and enables the network to achieve high feature extraction accuracy while reducing overall computational complexity compared to processing each feature map independently.
Solution Approach 2:
The patent employs dynamic feature fusion mechanisms that adaptively adjust the contribution of different feature maps based on their relevance to the current detection task. This dynamic approach allows the network to perform multiple convolution operations and feature fusions only when necessary, optimizing the balance between feature extraction accuracy and computational complexity according to the specific input characteristics.
Data Source
AI summary
A disaster smoke detection method includes: performing a first convolution operation on an input image to extract features to generate a primary feature map; performing enhancement processing on the primary feature map to obtain an enhanced feature map; performing multi-scale fusion on the enhanced feature map according to a second convolution operation to obtain a plurality of feature maps of different scales as high-level feature maps; respectively performing top-down feature fusion and bottom-up feature fusion on each high-level feature map at each scale according to a third convolution operation to correspondingly obtain a plurality of top-down fused feature maps and a plurality of bottom-up fused feature maps; fusing the top-down fused feature maps and the bottom-up fused feature maps to obtain a plurality of bidirectional cross-fused feature maps; and performing disaster and smoke detection on each of the bidirectional cross-fused feature maps.