Skip to main content

Table 1 Architectures of the proposed model

From: Plant diseases detection with low resolution data using nested skip connections

Blocks

Sub

Types

Output size

# Params

Initialization

–

\(\text {Conv2D } 3\times 3\)

\(31 \times 31 \times 64\)

1728

–

\(\text {Conv2D } 3\times 3\)

\(29 \times 29 \times 64\)

36,864

Compact block 1

Branch 1

\(\begin{bmatrix} \text {Prev. layer} \\ \text {Conv2D } 3\times 3 \end{bmatrix}\)

\(15 \times 15 \times 192\)

368,896

Branch 2

\(\text {Conv2D } 1\times 1\)

\(15 \times 15 \times 320\)

8192

Transition

–

\(\text {Conv2D } 3\times 3\)

\(8 \times 8 \times 128\)

369,920

Compact block 2

Branch 1

\(\begin{bmatrix} \text {Prev. layer} \\ \text {Conv2D } 3\times 3 \end{bmatrix}\)

\(4 \times 4 \times 384\)

885,248

Branch 2

\(\text {Conv2D } 1\times 1\)

\(4 \times 4 \times 640\)

32,768

Transition

–

\(\text {Conv2D } 3\times 3\)

\(2 \times 2 \times 256\)

1,477,120

Compact block 3

Branch 1

\(\begin{bmatrix} \text {Prev. layer} \\ \text {Conv2D } 3\times 3 \end{bmatrix}\)

\(2 \times 2 \times 768\)

3,539,968

Branch 2

\(\text {Conv2D } 1\times 1\)

\(1 \times 1 \times 1280\)

131,072

Transition

–

\(\text {Conv2D } 3\times 3\)

\(1 \times 1 \times 256\)

2,954,240

Classification layer

–

Glob. Av. pool.

\(256 \times 1\)

–

–

11 FC layer

\(11 \times 1\)

2827