https://orcid.org/0000-0002-7330-5118
Figures & data
Figure 1. (A) Penang map with a red square indicating the location of TAT-KM. (B) Site map of TAT-KM. (Google, n.d.).
Figure 2. Flow chart of the methodology.
Figure 3. (a) The confusion matrix, (b) Accuracy per epoch, (c) Loss per epoch for Model A.
Figure 4. (a) The confusion matrix, (b) Accuracy per epoch, (c) Loss per epoch for Model B.
Table 1. Number of training and test images used in Model A and Model B.
Figure 5. The average accuracy of Model A and Model B.
Figure 6. Number of test images with mean accuracy over 77% for each class.
Figure 7. Examples of training samples used in Model A. (I) Asian Dowitcher. (II) Brown-headed Gull. (III) Common Redshank. (IV) Whiskered Tern. (V) Great Egret. (VI) Little Egret. (VII) Pond Heron.
Figure 8. The difference in the appearance of the same species Asian dowitcher. (I) An example of a training image used in Model A. The morphological feature of the bird fits the description of the Asian dowitcher in the breeding season. (II) An example of a training image used in Model B. The brownish plumage tells that the bird is not in its breeding season.
Figure 9. Examples of training images of Model A that have complex backgrounds.
Figure 10. Examples of training images of Model B that have monotone backgrounds.
Figure 11. One of the training images in Model A. This image has 10 000 pixels (100 × 100).
Figure 12. One of the training images in Model A. This image has 24 megapixels (6000 × 4000).
Figure 13. One of the training images in Model B. This image has 2 megapixels (1607 × 1177).
Figure 14. One of the training images in Model B. This image has 12 megapixels (4032 × 3024).
