Data-Centric Preprocessing Outperforms Loss Modifications for Hard-Class Plant Disease Classification Using MobileNetV3
DOI:
https://doi.org/10.21831/elinvo.v11i1.95319Keywords:
data‑centric pipeline, image quality assessment, plant disease detection, targeted augmentation, lightweight CNNAbstract
Data-centric approaches have gained attention in plant disease classification; however, a systematic evaluation of underperforming ‘hard classes’ remains limited. This study proposes a four phase pipeline comprising granular error diagnosis, no-reference image quality assessment, class-targeted augmentation, and an ablation study to improve hard-class robustness without increasing model complexity. Using the New Plant Disease dataset and a lightweight MobileNetV3-Small backbone, we first established a baseline. Based on this baseline performance, we identified 12 hard classes (defined as those with either F1 < 0.96 or recall < 0.96 on the baseline model) with a mean F1 of 0.9260. The optimal configuration (categorical cross-entropy, no class weighting, baseline head) raised the mean hard-class F1 to 0.9668, corresponding to an absolute improvement of +4.08 percentage points, while maintaining global test accuracy at 98.08%. A two-tier design with three random seeds confirmed robustness (mean hard-class F1 = 0.9652 ± 0.0020). Unexpectedly, after data enrichment, inverse frequency class weighting degraded hard-class F1 (to 0.9621), and the default focal loss parameters offered no additional benefit over plain cross-entropy (0.9647). The MobileNetV3-Small showed comparable performance to the heavier EfficientNetB0 under our evaluation protocol, with no statistically significant difference detected (p = 0.089; limited power due to n = 3). Tomato Target Spot (C35) remained the most persistent bottleneck (F1 ≈ 0.90). Future work includes explainable AI for error analysis and real‑field validation.
References
[1] A. T. Khan, S. M. Jensen, A. R. Khan, and S. Li, “Plant disease detection model for edge computing devices,” Front. Plant Sci., vol. 14, p. 1308528, Dec. 2023, doi: 10.3389/fpls.2023.1308528.
[2] J. Yang, G. Xu, M. Yang, and Z. Lin, “Lightweight wavelet-CNN tea leaf disease detection,” PLOS One, vol. 20, no. 5, p. e0323322, May 2025, doi: 10.1371/journal.pone.0323322.
[3] U. V. Nnamdi and V. Abolghasemi, “Optimised MobileNet for very lightweight and accurate plant leaf disease detection,” Sci. Rep., vol. 15, no. 1, p. 43690, Dec. 2025, doi: 10.1038/s41598-025-27393-z.
[4] J. Zhang, X. Yang, X. Fu, B. Wang, and H. Li, “LDL-MobileNetV3S: an enhanced lightweight MobileNetV3-small model for potato leaf disease diagnosis through multi-module fusion,” Front. Plant Sci., vol. 16, p. 1656731, Oct. 2025, doi: 10.3389/fpls.2025.1656731.
[5] A. Y. Ashurov et al., “Enhancing plant disease detection through deep learning: a Depthwise CNN with squeeze and excitation integration and residual skip connections,” Front. Plant Sci., vol. 15, p. 1505857, Jan. 2025, doi: 10.3389/fpls.2024.1505857.
[6] N. Bhatt, N. Bhatt, P. Prajapati, V. Sorathiya, S. Alshathri, and W. El-Shafai, “A Data-Centric Approach to improve performance of deep learning models,” Sci. Rep., vol. 14, no. 1, p. 22329, Sep. 2024, doi: 10.1038/s41598-024-73643-x.
[7] D. Schaudt et al., “Augmentation strategies for an imbalanced learning problem on a novel COVID-19 severity dataset,” Sci. Rep., vol. 13, no. 1, p. 18299, Oct. 2023, doi: 10.1038/s41598-023-45532-2.
[8] R. Escobar Díaz Guerrero, L. Carvalho, T. Bocklitz, J. Popp, and J. L. Oliveira, “A Data Augmentation Methodology to Reduce the Class Imbalance in Histopathology Images,” J. Imaging Inform. Med., vol. 37, no. 4, pp. 1767–1782, Mar. 2024, doi: 10.1007/s10278-024-01018-9.
[9] D. Dablain, C. Bellinger, B. Krawczyk, and N. Chawla, “Efficient Augmentation for Imbalanced Deep Learning,” Oct. 17, 2022, arXiv: arXiv:2207.06080. doi: 10.48550/arXiv.2207.06080.
[10] Y. Yang, C. Liu, H. Wu, and D. Yu, “A quality assessment algorithm for no-reference images based on transfer learning,” PeerJ Comput. Sci., vol. 11, p. e2654, Jan. 2025, doi: 10.7717/peerj-cs.2654.
[11] Q.-H. Miao et al., “Deep learning-based no-reference quality assessment of anterior segment ultrasound biomicroscopy panoramic images,” Quant. Imaging Med. Surg., vol. 15, no. 11, pp. 11219–11236, Nov. 2025, doi: 10.21037/qims-2025-592.
[12] J. Singh et al., “Batch-balanced focal loss: a hybrid solution to class imbalance in deep learning,” J. Med. Imaging, vol. 10, no. 05, Jun. 2023, doi: 10.1117/1.JMI.10.5.051809.
[13] M. S. Hossain, J. M. Betts, and A. P. Paplinski, “Dual Focal Loss to address class imbalance in semantic segmentation,” Neurocomputing, vol. 462, pp. 69–87, Oct. 2021, doi: 10.1016/j.neucom.2021.07.055.
[14] Z. Al Sahili and M. Awad, “The power of transfer learning in agricultural applications: AgriNet,” Front. Plant Sci., vol. 13, p. 992700, Dec. 2022, doi: 10.3389/fpls.2022.992700.
[15] A. Ahmad, D. Saraswat, V. Aggarwal, A. Etienne, and B. Hancock, “Performance of deep learning models for classifying and detecting common weeds in corn and soybean production systems,” Comput. Electron. Agric., vol. 184, p. 106081, May 2021, doi: 10.1016/j.compag.2021.106081.
[16] X. Sun, G. Li, P. Qu, X. Xie, X. Pan, and W. Zhang, “Research on plant disease identification based on CNN,” Cogn. Robot., vol. 2, pp. 155–163, 2022, doi: 10.1016/j.cogr.2022.07.001.
[17] G. A. Sandag, F. M. Tombeng, S. R. Vridvly, J. Waworundeng, and W. Mokodaser, “CornNet: Implementation of Transfer Learning for Disease Classification in Corn Plants Based on Web Application,” CogITo Smart J., vol. 11, no. 1, pp. 167–181, Jun. 2025, doi: https://doi.org/10.31154/cogito.v11i1.992.167-181.
[18] B. Priambodo, A. Fadlil, and S. Sunardi, “Perbandingan CNN untuk Deteksi Penyakit Daun Tanaman New Plant Disease Berbasis Cloud Computing,” J. Inform. Teknol. Dan Sains, vol. 7, no. 4, 2025, doi: https://doi.org/10.51401/jinteks.v7i4.6782.
[19] N. Ullah et al., “An effective approach for plant leaf diseases classification based on a novel DeepPlantNet deep learning model,” Front. Plant Sci., vol. 14, p. 1212747, Oct. 2023, doi: 10.3389/fpls.2023.1212747.
[20] J. S. Aguilar-Ruiz and M. Michalak, “Classification performance assessment for imbalanced multiclass data,” Sci. Rep., vol. 14, no. 1, p. 10759, May 2024, doi: 10.1038/s41598-024-61365-z.
[21] Y. Toda and F. Okura, “How Convolutional Neural Networks Diagnose Plant Disease,” Plant Phenomics, vol. 2019, p. 9237136, 2019, doi: 10.34133/2019/9237136.
[22] S. Pertuz, D. Puig, and M. A. Garcia, “Analysis of focus measure operators for shape-from-focus,” Pattern Recognit., vol. 46, no. 5, pp. 1415–1432, May 2013, doi: 10.1016/j.patcog.2012.11.011.
[23] R. C. Gonzalez and R. E. Woods, Digital image processing, Fourth edition, Global edition. New York: Pearson, 2017.
[24] P. Gao, Z. Li, and H. Zhang, “Thermodynamics-Based Evaluation of Various Improved Shannon Entropies for Configurational Information of Gray-Level Images,” Entropy, vol. 20, no. 1, p. 19, Jan. 2018, doi: 10.3390/e20010019.
[25] M. Yeung, E. Sala, C.-B. Schönlieb, and L. Rundo, “Unified Focal loss: Generalising Dice and cross entropy-based losses to handle class imbalanced medical image segmentation,” Comput. Med. Imaging Graph., vol. 95, p. 102026, Jan. 2022, doi: 10.1016/j.compmedimag.2021.102026.
[26] T.-Y. Lin, P. Goyal, R. Girshick, K. He, and P. Dollár, “Focal Loss for Dense Object Detection,” in 2017 IEEE International Conference on Computer Vision (ICCV), 2017, pp. 2999–3007. doi: 10.1109/ICCV.2017.324.
[27] Y. Cui, M. Jia, T.-Y. Lin, Y. Song, and S. Belongie, “Class-Balanced Loss Based on Effective Number of Samples,” in 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Long Beach, CA, USA: IEEE, Jun. 2019, pp. 9260–9269. doi: 10.1109/CVPR.2019.00949.
[28] N. Liu, J. Wang, Y. Zhu, L. Wan, and Q. Li, “Improving imbalance classification via ensemble learning based on two-stage learning,” Front. Comput. Neurosci., vol. 17, p. 1296897, Jan. 2024, doi: 10.3389/fncom.2023.1296897.
[29] G. Habib and S. Qureshi, “GAPCNN with HyPar: Global Average Pooling convolutional neural network with novel NNLU activation function and HYBRID parallelism,” Front. Comput. Neurosci., vol. 16, p. 1004988, Nov. 2022, doi: 10.3389/fncom.2022.1004988.
[30] S. Bhattarai, “New Plant Disease Dataset.” 2018. Accessed: Aug. 16, 2025. [Online]. Available: https://www.kaggle.com/datasets/vipoooool/new-plant-diseases-dataset
[31] S. P. Mohanty, D. P. Hughes, and M. Salathé, “Using Deep Learning for Image-Based Plant Disease Detection,” Front. Plant Sci., vol. 7, p. 1419, Sep. 2016, doi: 10.3389/fpls.2016.01419.
[32] S. P. Mohanty, “Dataset of diseased plant leaf images and corresponding labels.” Accessed: Aug. 16, 2025. [Online]. Available: https://github.com/spMohanty/PlantVillage-Dataset
[33] D. Hughes and M. Salathé, “An open access repository of images on plant health to enable the development of mobile disease diagnostics,” Nov. 25, 2015. doi: https://arxiv.org/abs/1511.08060v2.
[34] S. M. Hassan, A. K. Maji, M. Jasiński, Z. Leonowicz, and E. Jasińska, “Identification of Plant-Leaf Diseases Using CNN and Transfer-Learning Approach,” Electronics, vol. 10, no. 12, p. 1388, Jun. 2021, doi: 10.3390/electronics10121388.
[35] S. B. Imanulloh, A. R. Muslikh, and D. R. I. M. Setiadi, “Plant Diseases Classification based Leaves Image using Convolutional Neural Network,” J. Comput. Theor. Appl., vol. 1, no. 1, pp. 1–10, Aug. 2023, doi: 10.33633/jcta.v1i1.8877.
Downloads
Published
How to Cite
Issue
Section
Citation Check
License
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
The article published in ELINVO became ELINVO's right in publication.
This work by ELINVO is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
