Attention-Driven Dual Optimization Framework for Accurate and Generalizable Maize Leaf Disease Classification

Abstract

This paper, proposes a Dual-Optimization Convolutional Neural Network to improve the accuracy and generalization of maize leaf disease classification. To address the constraints of predefined CNN architectures and single-algorithm hyperparameter consistence optimization, we combine two meta-heuristics optimizers with an architectural attention layer. In the proposed dual-optimization framework, PSO determines the number of convolutional filters in a dynamic manner, and BO fine-tunes the size of dense layers and global learning rate. The intermediate hybrid CNN structure is then instantiated with the best found FCSs and two other models are inspected: the hyper parameter hybrid model and the improved ancestral squeeze and excitation-based hybrid model that introduces modified SE attention blocks for concentrating on more discriminative, lesion-related zones. All models were trained with focal loss, dynamic L2 regularization, and Learning Rate Scheduler to obtain solid and stable class imbalance handling. Competitive experiments over the test set show that the hybrid model leads to higher classification accuracy and generalization, which confirms that combined optimization on hyperparameters and attention-driven learning directly sup-ports establishing a cutting-edge diagnostic model in our proposed co-design strategy. The proposed model achieved an accura-cy of approximately 89%, outperforming the baseline CNN, PSO, and BO models, thus validating the effectiveness of the dual optimization framework

Introduction

Maize is one of the world's most important staple crops, but its productivity is significantly affected by foliar diseases such as Northern Leaf Blight (NLB), Southern Leaf Blight (SLB), Gray Leaf Spot (GLS), Common Rust, and Maize Dwarf Mosaic Virus (MDMV). Traditional disease diagnosis through manual inspection is time-consuming, subjective, and unreliable, especially in large farms and during the early stages of infection. Recent advances in deep learning, particularly Convolutional Neural Networks (CNNs), have enabled accurate image-based disease detection using smartphone and camera images.

Existing CNN architectures such as ResNet, DenseNet, Inception, MobileNet, and SqueezeNet achieve high classification accuracy, while attention mechanisms, knowledge distillation, and hybrid CNN–Transformer models further improve feature extraction and generalization. However, current approaches still face challenges including class imbalance, overfitting, varying field conditions, complex backgrounds, and the computational limitations of edge devices. Models trained on laboratory datasets often perform poorly in real agricultural environments.

The literature indicates that lightweight, attention-guided architectures combined with adaptive optimization offer the best balance between accuracy and efficiency. Motivated by these research gaps, the proposed study introduces a Dual-Optimized Attention-Guided CNN with Adaptive Regularization for maize leaf disease classification. The framework integrates Particle Swarm Optimization (PSO), Bayesian Optimization (BO), Squeeze-and-Excitation (SE) attention, Focal Loss, and adaptive L2 regularization to improve feature extraction, handle class imbalance, reduce overfitting, and enhance robustness under real-world field conditions.

The methodology employs a hybrid optimization strategy in which PSO optimizes convolutional filter configurations, while BO fine-tunes continuous hyperparameters such as learning rate, dropout, and dense-layer units. Focal Loss addresses imbalanced datasets, and SE attention highlights disease-relevant regions to improve classification accuracy. Experimental comparisons with conventional CNN, PSO-CNN, and BO-CNN models demonstrate that the proposed hybrid CNN achieves superior accuracy, precision, recall, and F1-score, making it suitable for practical deployment in precision agriculture and edge-based disease monitoring systems.

Conclusion

The developed Dual Optimized CNN framework?combined PSO (Particle Swarm Optimization) and BO techniques showed better performance in maize leaf disease classification. The proposed framework combines the global search ability of PSO in feature extraction?step with more accuracy-oriented fine-tuning role of BO in classification head to be able to obtain close-to-optimal hy-perparameter tuning settings which resulted in enhancing both performance and generalization performance. The Squeeze-and-Excitation (SE) attention mechanisms were also adopted to introduce dynamic focusing capability of the important?diseased areas. At the same time, Dynamic L2 Regularization, Focal Loss and the Learning Rate Scheduler reinforced its?resistance to overfitting and class imbalance. Comparative study made sure that SE-Attn Hybrid CNN model had statistically better or equivalent prediction than the best single optimizers, demonstrating the effectiveness and?efficiency of hybrid optimization approach. In this paper, we propose a new efficient co-design strategy that combines hybrid optimization and attention driven learning for?the state-of-the-art intelligent crop disease diagnosis.

References

[1] M. A. Kiruthika, A. Gladston, and H. K. Nehemiah, “Lightweight dual-stage feature refinement for black gram leaf disease classification using ConViTSE,” Sci. Rep., vol. 15, no. 1, pp. 1–16, Nov. 2025, doi: 10.1038/s41598-025-22847-w. [2] J. Gao et al., “Integration of UAV and Remote Sensing Data for Early Diagnosis and Severity Mapping of Diseases in Maize Crop Through Deep Learning and Reinforcement Learning,” Remote Sens., vol. 17, no. 23, p. 3787, Nov. 2025, doi: 10.3390/rs17233787. [3] N. Parashar et al., “Enhanced residual-attention deep neural network for disease classification in maize leaf images,” Sci. Rep., vol. 15, no. 1, p. 29452, Aug. 2025, doi: 10.1038/s41598-025-14726-1. [4] K. Mookkandi et al., “A Robust Lightweight Vision Transformer for Classification of Crop Diseases,” AgriEngineering, vol. 7, no. 8, p. 268, Aug. 2025, doi: 10.3390/agriengineering7080268. [5] X. Gao et al., “DSTANet: A Lightweight and High-Precision Network for Fine-Grained and Early Identification of Maize Leaf Diseases in Field Environments,” Sensors, vol. 25, no. 16, p. 4954, Aug. 2025, doi: 10.3390/s25164954. [6] A. O. Akanmu et al., “Application of bioinoculants in combating Striga hermonthica (Del.) Benth infestation in maize,” BMC Microbiol., vol. 25, no. 1, pp. 1–14, Aug. 2025, doi: 10.1186/s12866-025-04264-1. [7] K. Mookkandi and M. K. Nath, “Robust Deep Neural Network for Classification of Diseases from Paddy Fields,” AgriEngineering, vol. 7, no. 7, p. 205, Jul. 2025, doi: 10.3390/agriengineering7070205. [8] H. E. Bakbak, C. Bal?m, and A. Savran, “A Hybrid Machine Learning Approach for Detecting and Assessing Zyginidia pullula Damage in Maize Leaves,” Appl. Sci., vol. 15, no. 10, p. 5432, May 2025, doi: 10.3390/app15105432. [9] C. Subbarayudu and M. Kubendiran, “Segmentation-based lightweight multi-class classification model for crop disease detection, classification, and severity assessment using DCNN,” PLoS One, vol. 20, no. 5, p. e0322705, May 2025, doi: 10.1371/journal.pone.0322705. [10] G. T. Askale et al., “Mobile based deep CNN model for maize leaf disease detection and classification,” Plant Methods, vol. 21, no. 1, pp. 1–15, May 2025, doi: 10.1186/s13007-025-01386-5. [11] M. H. Lye, M. F. A. Fauzi, and L. K. Ming, “Maize Leaf Disease Identification with Large and Lightweight Convolutional Neural Models,” Int. J. Informat. Visual., vol. 8, no. 2, pp. 1–7, Mar. 2025, doi: 10.62527/IJIV.8.2.1. [12] S. M. Jibril et al., “Microbiome Analysis of Area in Proximity to White Spot Lesions Reveals More Harmful Plant Pathogens in Maize,” Biomolecules, vol. 15, no. 2, p. 252, Feb. 2025, doi: 10.3390/biom15020252. [13] D. T. Weldeslasie et al., “A statistically validated stacking ensemble of CNNs and vision transformer for robust maize disease classification,” Discov. Artif. Intell., vol. 5, no. 1, pp. 1–18, Jan. 2025, doi: 10.1007/s44163-025-00548-7. [14] X. Jia et al., “Monitoring Maize Leaf Spot Disease Using Multi-Source UAV Imagery,” Drones, vol. 7, no. 11, p. 650, Oct. 2023, doi: 10.3390/drones7110650. [15] M. A. Haque et al., “Deep learning-based approach for identification of diseases of maize crop,” Sci. Rep., vol. 12, no. 1, p. 6300, Apr. 2022, doi: 10.1038/s41598-022-10229-3. [16] H. Liu et al., “Research On Maize Disease Identification Methods In Complex Environments Based On Cascade Networks And Two-Stage Transfer Learning,” Sci. Rep., vol. 12, no. 1, p. 18944, Nov. 2022, doi: 10.1038/s41598-022-23352-5. [17] L. Luo, Q. Chang, Q. Wang, and Y. Huang, “Identification and Severity Monitoring of Maize Dwarf Mosaic Virus Infection Based on Hyperspectral Measurements,” Remote Sens., vol. 13, no. 22, p. 4560, Nov. 2021, doi: 10.3390/rs13224560. [18] Y. Huangfu et al., “Genome-wide identification of PTI1 family in Setaria italica and salinity-responsive functional analysis of SiPTI1-5,” BMC Plant Biol., vol. 21, no. 1, p. 313, Jul. 2021, doi: 10.1186/s12870-021-03080-8. [19] Z. Cui et al., “Denser Markers and Advanced Statistical Method Identified More Genetic Loci Associated with Husk Traits in Maize,” Sci. Rep., vol. 10, no. 1, p. 8647, May 2020, doi: 10.1038/s41598-020-65164-8. [20] M. Kannan, I. Ismail, and H. Bunawan, “Maize Dwarf Mosaic Virus: From Genome to Disease Management,” Viruses, vol. 10, no. 9, p. 492, Sep. 2018, doi: 10.3390/v10090492. [21] M. J. Wamaitha et al., “Metagenomic analysis of viruses associated with maize lethal necrosis in Kenya,” Virol. J., vol. 15, no. 1, p. 90, May 2018, doi: 10.1186/s12985-018-0999-2. [22] Z. Ahmad, J. Wu, L. Chen, and W. Dong, “Isolated Bacillus subtilis strain 330-2 and its antagonistic genes identified by the removing PCR,” Sci. Rep., vol. 7, no. 1, p. 1777, May 2017, 2017, doi: 10.1038/s41598-017-01940-9. [23] K. Magade and A. Sharma, “Significant role of IoT in Cyber-Physical Systems, Context Awareness, and Ambient Intelligence,” The Next Generation Innovation in IoT and Cloud Computing with Applications. 2024. doi: 10.1201/9781003406723-2. [24] P. Garg and A. Sharma, “A distributed algorithm for local decision of cluster heads in wireless sensor networks,” IEEE International Conference on Power, Control, Signals and Instrumentation Engineering, ICPCSI 2017. pp. 2411–2415, 2018. doi: 10.1109/ICPCSI.2017.8392150. [25] A. Sharma and A. Sharma, “KNN-DBSCAN: Using k-nearest neighbor information for parameter-free density based clustering,” 2017 International Conference on Intelligent Computing, Instrumentation and Control Technologies, ICICICT 2017, vol. 2018-January. pp. 787–792, 2017. doi: 10.1109/ICICICT1.2017.8342664. [26] P. Sharma and A. Sharma, “Online K-means clustering with adaptive dual cost functions,” 2017 International Conference on Intelligent Computing, Instrumentation and Control Technologies, ICICICT 2017, vol. 2018-January. pp. 793–799, 2017. doi: 10.1109/ICICICT1.2017.8342665. [27] D. Dadhich and A. Sharma, “Improved Online k-Means Algorithm,” presented at the 11th INDIACom; 4th International Conference on “Computing for Sustainable Global Development”, INDIACom 2017, 2017, pp. 1286–1292. [Online]. Available: https://www.scopus.com/inward/record.uri?eid=2-s2.0-105010812740&partnerID=40&md5=a15d59a1c922fa1e55533e0e79ce5a67 [28] N. Mittal and A. Sharma, “Energy-Efficient Fuzzy-Logic Based Cluster Selection for Heterogeneous Mobile Nodes in Wireless Sensor Networks,” presented at the 11th INDIACom; 4th International Conference on “Computing for Sustainable Global Development”, INDIACom 2017, 2017, pp. 876–881. [Online]. Available: https://www.scopus.com/inward/record.uri?eid=2-s2.0-105010821301&partnerID=40&md5=7be239921a1c966e52084f6c49f5d9df [29] Ananda S. Paymode, Vandana B. Malode, Transfer Learning for Multi-Crop Leaf Disease Image Classification using Convolutional Neural Network VGG, Artificial Intelligence in Agriculture, Volume 6, 2022, Pages 23-33,ISSN 2589-7217, https://doi.org/10.1016/j.aiia.2021.12.002.

Copyright

Copyright © 2026 Rohit Maheshwari, Awamit Kumar, Amit Sharma . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.