Table 1 Abbreviations and symbols.
From: DEENet: an edge-enhanced CNN–Transformer dual-encoder model for steel surface defect detection
Symbol | Explain | |
|---|---|---|
Model | DEENet | This paper proposes an edge enhancement CNN–Transformer dual encoder model |
AKConv | Adaptive-Kernel Convolution | |
DCF | Dual Channel Fusion | |
C2f_EEM | Edge-Enhancement Module | |
CBAM | Convolutional Block Attention Module | |
ViT | Vision Transformer | |
Evaluation indicators | mAP | mean Average Precision |
F1-score | Harmonic mean of precision and recall | |
FPS | Frames Per Second | |
GFLOPs | Measuring the computational complexity and resource consumption of the model | |
Mathematical symbols | \(C_{i}\) \(T_{i}\) | These represent feature maps from the CNN branch and the Transformer branch, respectively |
\(Q,K,V\) | Query, Key, and Value in Attention Mechanisms | |
\(\sigma\) | Sigmoid activation function | |
\(F_{a}\)\(F_{b}\) | These represent the shallow and deep features in the edge enhancement module, respectively | |
\(\varphi (Z)\) | High-frequency feature maps are used to extract edge information through differential operations | |
\(\oplus\) | Element addition in residual join |
