Research Article
[Retracted] Identification of Navel Orange Diseases and Pests Based on the Fusion of DenseNet and Self-Attention Mechanism
Table 3
Main parameters of DCPSNET.
| | Layer name | Type | Input shape | Stride | Output size | Maps | Repeated times |
| | Input layer | Input layer | 224 × 224 × 3 | — | 224 × 224 | — | 1 | | First layer | Conv2d | 224 × 224 × 3 | 2 | 112 × 112 | 64 | 1 | | DCSAM block1 | SeparableConv2D | 112 × 112 × 64 | 2 | 56 × 56 | 160 | 6 | | CSAM | Shortcutconnection | 56 × 56 × 160 | 1 | 56 × 56 | 160 | 1 | | PSAM | Shortcutconnection | 56 × 56 × 160 | 1 | 56 × 56 | 160 | 1 | | Transform1 | Conv2D + Avpooling | 56 × 56 × 160 | 1 | 56 × 56 | 80 | 1 | | DCSAM block2 | SeparableConv2D | 56 × 56 × 80 | 2 | 28 × 28 | 272 | 12 | | CSAM | Shortcutconnection | 28 × 28 × 272 | 1 | 28 × 28 | 272 | 1 | | PSAM | Shortcutconnection | 28 × 28 × 272 | 1 | 28 × 28 | 272 | 1 | | Transform2 | Conv2D + Avpooling | 28 × 28 × 272 | 1 | 28 × 28 | 136 | 1 | | DCSAM block3 | SeparableConv2D | 28 × 28 × 136 | 2 | 14 × 14 | 520 | 24 | | CSAM | Shortcutconnection | 14 × 14 × 520 | 1 | 14 × 14 | 520 | 1 | | PSAM | Shortcutconnection | 14 × 14 × 520 | 1 | 14 × 14 | 520 | 1 | | Transform3 | Conv2D + Avpooling | 14 × 14 × 520 | 1 | 14 × 14 | 260 | 1 | | DCSAM block4 | SeparableConv2D | 14 × 14 × 260 | 2 | 7 × 7 | 516 | 16 | | CSAM | Shortcutconnection | 7 × 7 × 516 | 1 | 7 × 7 | 516 | 1 | | PSAM | Shortcutconnection | 7 × 7 × 516 | 1 | 7 × 7 | 516 | 1 | | Final_conv | BN + ReLU | 7 × 7 × 258 | 7 | 1 × 1 | 7 | 1 | | Linear | Classification | 1 × 1 × 7 | 1 | 7 | 1 | 1 |
|
|
