Fusi Fitur Deep dan Manual Berbasis Masker Warna dengan Seleksi F-ANOVA Klasifikasi Penyakit Padi

Agung Ramadhanu

doi:10.35134/komtekinfo.v11i3.718

Authors

Agung Ramadhanu Universitas Putra Indonesia YPTK Padang

DOI:

https://doi.org/10.35134/komtekinfo.v11i3.718

Keywords:

Rice Leaf Disease Classification, Feature Fusion, EfficientNet-B0, Data Leakage, Ablation Study

Abstract

Studies on rice leaf disease classification frequently report accuracy above 99%, yet such figures can be inflated when duplicate images are shared between training and test partitions. This study evaluates a dual-branch hybrid pipeline for rice leaf disease classification under a leakage-controlled evaluation protocol. A public Kaggle dataset of 7,421 images was deduplicated by exact perceptual-hash matching, leaving 4,410 unique images in five classes. Deep features (1,280 dimensions) from an ImageNet-pretrained EfficientNet-B0 were combined with 73 handcrafted descriptors computed from colour-segmentation masks (colour moments, masked GLCM, LBP, and mask morphometry). The proposed scenario (E5) keeps the 128 features with the highest ANOVA F-scores, with scaling and selection fitted on training folds only. Five ablation scenarios (E1–E5) were evaluated with SVM-RBF and LightGBM using repeated stratified 5-fold cross-validation with three seeds (15 test folds) and Wilcoxon signed-rank tests. With SVM-RBF, E5 reached 99.64% ± 0.18% accuracy and 99.47% ± 0.28% macro-F1 while using 90.5% fewer features than naive concatenation (1,353 features). E5 was better than handcrafted features alone (p = 0.0067; Holm-adjusted p = 0.034), but this was not significant under a resampled t-test corrected for fold dependence (p = 0.17), and E5 was not significantly different from deep features alone or naive concatenation (p > 0.05). With LightGBM, E5 was significantly worse than E1, E3, and E4. Visual inspection of one image per class showed that the colour-based masks do not reliably isolate visible lesions, so the handcrafted features should not be read as lesion-specific. The compact representation is therefore competitive rather than superior, and its behaviour depends on the classifier. All results are in-distribution; field generalisation was not tested.

References

S. Lamba, V. Kukreja, J. Rashid, T. R. Gadekallu, J. Kim, A. Baliyan, D. Gupta, and S. Saini, “A novel fine-tuned deep-learning-based multi-class classifier for severity of paddy leaf diseases,” Front. Plant Sci., vol. 14, Art. no. 1234067, 2023, doi: 10.3389/fpls.2023.1234067.

C. Subbarayudu and M. Kubendiran, “An automated hybrid deep learning framework for paddy leaf disease identification and classification,” Sci. Rep., vol. 15, Art. no. 26873, 2025, doi: 10.1038/s41598-025-08071-6.

M. H. Ishaq and C. Direkoğlu, “A deep learning based fusion model for plant leaf disease classification,” in Proc. 7th Int. Congr. Human-Computer Interaction, Optimization and Robotic Applications (ICHORA), Ankara, Türkiye, May 2025, doi: 10.1109/ICHORA65333.2025.11017095.

M. A. Uddin and J. B. Joolee, “Fusion of deep learned and handcrafted features for paddy disease recognition,” in Proc. 2024 Int. Conf. Artif. Intell., Metaverse Cybersecur. (ICAMAC), Dubai, UAE, Oct. 2024, Art. no. 10829316, doi: 10.1109/ICAMAC62387.2024.10829316.

M. Aggarwal, V. Khullar, N. Goyal, A. Singh, A. Tolba, E. B. Thompson, and S. Kumar, “Pre-trained deep neural network-based features selection supported machine learning for rice leaf disease classification,” Agriculture, vol. 13, no. 5, Art. no. 936, 2023, doi: 10.3390/agriculture13050936.

M. Jelali, “Deep learning networks-based tomato disease and pest detection: a first review of research studies using real field datasets,” Front. Plant Sci., vol. 15, Art. no. 1493322, 2024, doi: 10.3389/fpls.2024.1493322.

M. A. Noyan, “Uncovering bias in the PlantVillage dataset,” arXiv:2206.04374, 2022.

S. Kapoor and A. Narayanan, “Leakage and the reproducibility crisis in machine-learning-based science,” Patterns, vol. 4, no. 9, Art. no. 100804, 2023, doi: 10.1016/j.patter.2023.100804.

B. Barz and J. Denzler, “Do we train on test data? Purging CIFAR of near-duplicates,” J. Imaging, vol. 6, no. 6, Art. no. 41, 2020, doi: 10.3390/jimaging6060041.

C. Ambroise and G. J. McLachlan, “Selection bias in gene extraction on the basis of microarray gene-expression data,” Proc. Natl. Acad. Sci. U.S.A., vol. 99, no. 10, pp. 6562–6566, 2002, doi: 10.1073/pnas.102102699.

A. Ahmad, A. El Gamal, and D. Saraswat, “Toward generalization of deep learning-based plant disease identification under controlled and field conditions,” IEEE Access, vol. 11, pp. 9042–9057, 2023, doi: 10.1109/ACCESS.2023.3240100.

“Rice Leaf Disease Detection Dataset,” Kaggle, pengguna shams069. [Online]. Available: https://www.kaggle.com/datasets/shams069/rice-leaf-disease-detection-dataset

M. Tan and Q. Le, “EfficientNet: Rethinking model scaling for convolutional neural networks,” in Proc. 36th Int. Conf. Mach. Learn. (ICML), PMLR 97, 2019, pp. 6105–6114.

R. M. Haralick, K. Shanmugam, and I. Dinstein, “Textural features for image classification,” IEEE Trans. Syst., Man, Cybern., vol. SMC-3, no. 6, pp. 610–621, 1973, doi: 10.1109/TSMC.1973.4309314.

F. Pedregosa et al., “Scikit-learn: Machine learning in Python,” J. Mach. Learn. Res., vol. 12, pp. 2825–2830, 2011.

C. Cortes and V. Vapnik, “Support-vector networks,” Mach. Learn., vol. 20, pp. 273–297, 1995, doi: 10.1007/BF00994018.

G. Ke et al., “LightGBM: A highly efficient gradient boosting decision tree,” in Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 30, 2017, pp. 3146–3154.

F. Wilcoxon, “Individual comparisons by ranking methods,” Biometrics Bull., vol. 1, no. 6, pp. 80–83, 1945, doi: 10.2307/3001968.

S. Holm, “A simple sequentially rejective multiple test procedure,” Scand. J. Stat., vol. 6, no. 2, pp. 65–70, 1979.

C. Nadeau and Y. Bengio, “Inference for the generalization error,” Mach. Learn., vol. 52, no. 3, pp. 239–281, 2003, doi: 10.1023/A:1024068626366.

R. R. Bouckaert and E. Frank, “Evaluating the replicability of significance tests for comparing learning algorithms,” in Advances in Knowledge Discovery and Data Mining (PAKDD), Lecture Notes in Computer Science, 2004, pp. 3–12, doi: 10.1007/978-3-540-24775-3_3.

R. Maryani and [Nama Dosen Pembimbing], “Ilusi keterjelasan pada kuantifikasi penjajaran Grad-CAM: studi kasus masker segmentasi warna otomatis pada citra penyakit padi,” [jurnal dan status: diisi penulis], 2026.

R. Maryani and [Nama Dosen Pembimbing], “Evaluasi pergeseran domain zero-shot dan latensi tahap klasifikasi pada pipeline hibrida fitur terseleksi untuk penyakit daun padi,” [jurnal dan status: diisi penulis], 2026.