1 of 1

Efficient Leaf Disease Classification and Segmentation using Midpoint Normalization Technique and Attention Mechanism

Enam Ahmed Taufik, Antara Firoz Parsa, Seraj Al Mahmud Mostafa

  • Image preprocessing is crucial in disease detection, as it enhances data quality for improved classification and segmentation.
  • Lightweight models are efficient but often underperform with Mid-point Normalization (MPN)-preprocessed images due to their limited feature extraction capacity.
  • Attention mechanisms help lightweight models focus on essential features, thereby improving accuracy with minimal computational cost.
  • U-Net segmentation performance improves with the integration of the SE block, aiding in the precise localization of diseased regions in plant leaf images.

BACKGROUND

Enam Ahmed Taufik1, Antara Firoz Parsa1, Seraj Al Mahmud Mostafa2 contact: 1{enam.ahmed.taufik, antara.firuz.parsa}@g.bracu.ac.bd, 2serajmostafa@umbc.edu

1Department of Computer Science and Engineering, BRAC University, Dhaka, Bangladesh, 2Department of Information Systems, University of Maryland, Baltimore County, USA

METHODS

  1. Classification Framework: A novel lightweight CNN architecture, SE-ConvNet, is proposed by integrating Squeeze-and-Excitation (SE) blocks to recalibrate feature maps via channel-wise attention.

2. Segmentation Pipeline: A customized U-Net model with encoder–decoder symmetry and skip connections is employed to preserve spatial detail. The architecture is enhanced with Batch Normalization, Conv2DTranspose layers for upsampling, and SE blocks to improve spatial coherence and boundary localization.

3. Loss Functions and Augmentation: For segmentation, a composite loss function combining Dice Coefficient and Intersection over Union (IoU) is used to enhance region-level accuracy:

where α,β ∈ [0,1]. A custom combined_generator augments training data.

CONCLUSION

  • The proposed SE-ConvNet,achieved a classification accuracy of 93%, outperforming standard lightweight models.
  • The integration of Squeeze-and-Excitation (SE) blocks improved feature representation, resulting in strong segmentation results (72.44% Dice, 58.54% IoU) with SE-U-Net.
  • These findings highlight the effectiveness of combining task-specific preprocessing and attention-based recalibration for building efficient, high-performing models suitable for low-resource applications.

[1] L. C. Ngugi, M. Abelwahab, and M. Abo-Zahhad, “Recent advances in image processing techniques for automated leaf pest and disease recognition–a review,” Information processing in agriculture, vol. 8, no. 1, pp. 27–51, 2021.

[2] M. Nawaz, T. Nazir, A. Javed, M. Masood, J. Rashid, J. Kim, and A. Hussain, “A robust deep learning approach for tomato plant leaf disease localization and classification,” Scientific reports, vol. 12, no. 1, p. 18568, 2022.

[3] W. Li, X. Yu, C. Chen, and Q. Gong, “Identification and localization of grape diseased leaf images captured by uav based on cnn,” Computers and Electronics in Agriculture, vol. 214, p. 108277, 2023.

[4] M. E. Pothen and M. L. Pai, “Detection of rice leaf diseases using image processing,” in 2020 fourth international conference on computing methodologies and communication (ICCMC). IEEE, 2020, pp. 424–430.

[5] M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “Mobilenetv2: Inverted residuals and linear bottlenecks,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018, pp. 4510–4520.

[6] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” arXiv preprint arXiv:1409.1556, 2014.

[7] K. Zuiderveld, “Contrast limited adaptive histogram equalization,” in Graphics gems IV. Academic Press Professional, Inc., 1994, pp. 474–485.

[8] J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7132–7141.

[9] O. Mzoughi and I. Yahiaoui, “Deep learning-based segmentation for disease identification,” Ecological Informatics, vol. 75, p. 102000, 2023.

[10] L. Divyanth, A. Ahmad, and D. Saraswat, “A two-stage deep-learning based segmentation model for crop disease quantification based on corn field imagery,” Smart Agricultural Technology, vol. 3, p. 100108, 2023.

[11] O.Ronneberger,P.Fischer,andT.Brox,“U-net:Convolutional networks for biomedical image segmentation,” in International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI). Springer, 2015, pp. 234–241.

[12] M. R. A. Rashid, M. M. Hossain, J. Biswas, and H. Majumder, “Betel leaf image dataset from bangladesh,” 2024. [Online]. Available: https://data.mendeley.com/datasets/g7fpgj57wc/2

[13] S. A. M. Mostafa, O. Faruque, C. Wang, J. Yue, S. Purushotham, and J. Wang, “gWaveNet: Classification of gravity waves from noisy satellite data using custom kernel integrated deep learning method,” in International Conference on Pattern Recognition. Springer, 2025, pp. 164–180.

[14] F. David and M. A. Mukunthan, “Betel leaf diseases classification using machine learning algorithm: A feasible approach,” Journal of Advanced Research in Applied Sciences and Engineering Technology, vol. 40, no. 1, p. 7486, 2024.

[15] G. Kumar, P. Sharma, and A. Gupta, “An integrated tea leaf diseases identification and retrieval model using machine learning and deep learning approach,” March 2024. [Online]. Available: http://dx.doi.org/10.2139/ssrn.4757484

[16] J. Ford, E. Sadgrove, and D. Paul, “Joint plant- spraypoint detector with convnext modules and hist- match normalization,” Precision Agriculture, vol. 26, p. 24, 2025.

[17] C. Hou, J. Zhuang, Y. Tang, Y. He, A. Miao, H. Huang, and S. Luo, “Recognition of early blight and late blight diseases on potato leaves based on graph cut segmentation,” Journal of Agriculture and Food Research, vol. 5, p. 100154, 2021.

[18] D. N. K. Pandiri, R. Murugan, and T. Goel, “Arm-unet: attention residual path modified unet model to segment the fungal pathogen diseases in potato leaves,” Signal, Image and Video Processing (SIViP), vol. 19, p. 80, 2025.

[19] K. Li, W. Gu, M. Xue, J. Xiao, D. Shi, and X. Wei, “Atten-adapter: A unified attention-based adapter for efficient tuning,” in 2023 IEEE International Conference on Image Processing (ICIP). IEEE, 2023.

[20] F. F. Niloy, M. A. Amin, A. A. Ali, and A. M. Rahman, “Attention toward neighbors: A context aware framework for high resolution image segmentation,” in 2021 IEEE International Conference on Image Processing (ICIP). IEEE, 2021, pp. 2279–2283.

[21] H. Luo, X. Zhang, and G. Guoy, “Convolutional attention model for restaurant recommendation with multi- view visual features,” in 2020 IEEE International Conference on Image Processing (ICIP). IEEE, 2020.

[22] Y. Song, L. Wu, Y. Zhao, P. Liu, R. Lv, and H. Ul- lah, “High-accuracy gesture recognition using mm- wave radar based on convolutional block attention module,” in 2023 IEEE International Conference on Image Processing (ICIP). IEEE, 2023, pp. 1485–1489.

[23] C. Li, Y. Tan, W. Chen, X. Luo, Y. Gao, X. Jia, and Z. Wang, “Attention unet++: A nested attention-aware u-net for liver ct image segmentation,” in 2020 IEEE international conference on image processing (ICIP). IEEE, 2020, pp. 345–349.

[24] R. Wu, X. Xiao, G. Hu, H. Zhao, H. Zhang, and Y. Peng, “Fibonet: A light-weight and efficient neural network for image segmentation,” in 2023 IEEE International Conference on Image Processing (ICIP). IEEE, 2023, pp. 1345–1349.

[25] S. A. M. Mostafa, C. Wang, J. Yue, Y. Hozumi, and J. Wang, “Enhancing Satellite Object Localization with Dilated Convolutions and Attention-aided Spatial Pooling,” arXiv preprint arXiv:2505.05599, 2025. [Online]. Available: https://arxiv.org/abs/2505.05599

REFERENCES

DISCUSSION

  • The SE-ConvNet demonstrated superior performance and compactness over conventional lightweight models, emphasizing the effectiveness of dataset-specific preprocessing and attention-based feature recalibration.
  • GradCAM++ visualizations confirmed that the model attends to disease-relevant regions, enhancing interpretability and validating the impact of integrated attention mechanisms.
  • While the incorporation of SE blocks improves performance, the use of a single attention head may limit complex feature representation.
  • Future work could explore multi-head attention and deeper architectures for greater robustness in diverse conditions.

a) Original Image

b) GradCam Colormap

Figure 03: Feature Visualization using GradCAM++

RESULT

Classification

Segmentation

Table 1: Classification Accuracies of Various Image Pre-processing Techniques.

Table 2: Performance Metrics of the Best Performing Model.

Table 3: Performance Metrics Comparison Between U-Net and U-Net with Attention Mechanism.

a) U-Net With SE-block

b) U-Net Without SE-block

Figure 02: Confusion Matrix of SE-ConvNet

Figure 01: Comparison of Losses between U-Net and U-Net with SE Block