Skip to main content

Table 3 Effect of CNN architecture on classification accuracy (cols. 2,3) and number of parameters per layer (cols. 4-9) in each architecture

From: DeepSort: deep convolutional networks for sorting haploid maize seeds

Method Train Test Conv1 Conv2 Full1 Full2 Output Total
Arch-1 1.000 0.968 1216 6416 786,624 18,528 194 812,978
Arch-2 1.000 0.968 608 1608 196,704 4656 98 203,674
Arch-3 0.992 0.941 304 404 49,200 1176 50 51,134
Arch-4 0.989 0.935 304 - 98,328 - 50 98,682
  1. Conv[1/2]: [first/second] convolution layers, Full[1/2] : [first/second] fully connected layers, output: final softmax layer
\