Table 1.
Summary data of split disease conditions.
Fig 1.
Samples of dataset classes.
Table 2.
Summary of datasets used in the study.
Fig 2.
Channel attention module.
Fig 3.
Spatial attention.
Fig 4.
Channel, spatial attention (CBAM).
Fig 5.
Learnable gated fusion CBAM (LGF-CBAM).
Fig 6.
The proposed CNN model.
Table 3.
Ablation study for optimizer selection.
Table 4.
Ablation study for activation function selection.
Fig 7.
Ablation study for batch size selection.
Fig 8.
Ablation study for pooling operation selection.
Table 5.
Selected hyperparameters for training the model.
Table 6.
Summary of feature extraction time for base models.
Table 7.
Training and inference time summary for base models.
Fig 9.
a: Accuracy for Pretrained Models. b. Model FPR for Pretrained Models. c. Model F1 score for Pretrained Models. d. Model PPV score for Pretrained Models. e. Model TPR score for Pretrained Models. f. Model Loss Score for Pretrained Models.
Table 8.
Summary of evaluation on the base pretrained CNN models.
Table 9.
Statistical significance test on the base pretrained CNN models.
Fig 10.
Training and validation accuracy for the cocoa disease GH dataset.
Fig 11.
Training and validation loss on the cocoa disease GH dataset.
Fig 12.
Confusion matrix for the test dataset on the cocoa disease GH dataset.
Table 10.
Summary of model performance metrics on the cocoa disease GH dataset.
Fig 13.
a: Grad-CAM visualizations of Healthy Cocoa Disease.b: Grad-CAM visualizations of Moni Cocoa Disease. c: Grad-CAM visualizations of Phyto Cocoa Disease.
Table 11.
Performance comparison of the proposed LGF-CBAM model with baseline CBAM-based architectures on the Cocoa Disease GH dataset.
Fig 14.
a: Accuracy for All Datasets. b: Model FPR for all datasets. c: Model F1 Score for All Datasets. d: Model PPV Score for All Datasets. e:Model TPR Score, All Datasets. f: Model Loss Score for All Datasets.
Table 12.
Summary of all dataset performance on the proposed model.
Table 13.
Cross-dataset performance.
Table 14.
Comparison of the proposed system with existing related systems.
Fig 15.
Comparison of cocoa diseases (YOLOv4) dataset.
Fig 16.
Comparison on cacao disease in davao dataset.
Fig 17.
Comparison on private datasets.