Evaluation of Saliency-Guided Foreground Extraction for Medicinal Plant Flower Classification on the PhulMedic Dataset
DOI:
10.59888/ajosh.v1i06.790Published:
2023-03-25Downloads
Abstract
Background: Medicinal plant flower image classification faces variations in background, lighting, and visual characteristics that can affect the feature representations of deep learning models, yet the benefit of background removal remains unclear.
Objective: This study evaluates whether Saliency-Guided Foreground Extraction improves MobileNetV3Small classification on the PhulMedic dataset and what computational cost it adds.
Method: The dataset contains 5,140 images of Crape Jasmine, Hibiscus, Madagascar Periwinkle, Marigold, and Wedelia, split with stratification at a 70:15:15 ratio. Original images were compared with foreground images produced by Static Saliency Spectral Residual, a threshold of 25, and morphological closing with a 7 × 7 elliptical kernel. An ImageNet-pretrained MobileNetV3Small with a frozen backbone and 224 × 224 inputs was trained under identical settings, and compared using the exact McNemar test.
Results: On 771 test images, the original condition achieved 100% accuracy and macro-F1, whereas the foreground condition achieved 99.61% for both, with three misclassifications; the difference was not statistically significant (p = 0.25). Foreground preprocessing plus training was 20.76% slower than baseline training.
Conclusion: Foreground extraction produced no detectable difference in performance while adding computational overhead. Because the baseline was saturated and only one split was used, this design cannot show whether foreground extraction helps.
Keywords:
PhulMedic flower classification saliency detection foreground extraction MobileNetV3SmallLicense
Copyright (c) 2023 Viandre Chika Baratama

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Authors who publish with this journal agree to the following terms:
- Authors retain copyright and grant the journal right of first publication with the work simultaneously licensed under a Creative Commons Attribution-ShareAlike 4.0 International. that allows others to share the work with an acknowledgement of the work's authorship and initial publication in this journal.
- Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgement of its initial publication in this journal.
- Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work.






