Table 2. SGNF optimized by GA.

Algorithm 2 SGNF Optimized by GA
1: Input the initial sample set {X}, and the K initial seeds are selected randomly.
2: Obtain the optimal K superpixel seeds to generate the initial SNGF with K neural tree using genetic algorithm.
3: Repeat
4: For each superpixel sample in {X} do
5: Generate neuron nj from {X} and traverse the SGNF to find a neuron nwin with the shortest distance to nj.
6: Connect nj to the SGNT.
If nwin is a leaf neuron in the current SGNT
Create a new neuron nj+1 and copy weight wwin to wj+1;
Connect nj and nj+1 to nwin as a child.
Else Connect nj to nwin as a child, update node weight.
7: End for.
8: Until all samples have been input into the SGNF.