Table 1 The parameter specifications for PCSN-Net model components.
From: A hybrid parallel convolutional spiking neural network for enhanced skin cancer detection
Layer/component | Parameter | Value/description |
|---|---|---|
Image pre-processing | Filter type | Medav filter |
Mean filter size | \(3 \times 3\) | |
Median filter size | \(3 \times 3\) | |
Segmentation | Model | DeepSegNet |
Encoder depth | 3 layers | |
Decoder depth | 3 layers | |
Fusion coefficient | RV coefficient | |
Image augmentation | Techniques | Mixup, CutMix, geometric transformation, colorspace transformation |
Rotation angle | \(0^{\circ }\) to \(90^{\circ }\) | |
Brightness range | \(\pm 0.2\) | |
Feature extraction | Texture features | Contrast, correlation, energy, homogeneity |
Statistical features | Mean, variance, skewness | |
PCNN model | Convolutional layers | Number of layers: 4 |
Filter size | \(3 \times 3\) | |
Strides | 1 | |
Activation function | ReLU | |
Pooling | Max pooling (size: \(2 \times 2\)) | |
PCSN-net layer | Fusion method | Regression via fractional calculus |
Statistical and textural features | 7 features | |
DSNN model | Neuron firing threshold | 0.5 |
Spike count learning rule | Spike vector quantization | |
Synaptic weights initialization | Xavier initialization | |
Spiking activation | Spiking ReLU |
