Fig 1.
Examples of images captured from sticky trap.
(a) trap with BPH and (b) trap with no BPH.
Fig 2.
Examples of cropped images from sticky pad.
(a) Positive patched containing a brown planthopper and (b) negative patches (benign).
Table 1.
Statistics of constructed datasets.
Fig 3.
Illustration of the PENYEK classification pipeline.
Fig 4.
Different types of filtering are applied on captured BP images.
These include outline, fill hole, skeletonize, distance map, watershed and Voronoi.
Table 2.
Details of a combination of system components.
Table 3.
Details of CNN architectures for BN classification on sticky pad datasets.
Fig 5.
VGG architecture.
Table 4.
Details of empirically evaluated CNN structures for BPH classification on sticky pad images.
5 cross-validation (CV).
Table 5.
5 CV.
Table 6.
Effectiveness of data augmentation.
5 CV.
Table 7.
Average classification accuracy and AUC of applied methods for BPH classification.
Fig 6.
Classification results of 12 test images fed into VGG16.
The image size is 244x244 in RGB image. BPH and benign insect pests are classified using the PENYEK pipeline at the specified confidence percentage.