Answers are grounded in this paper and cite their sources.
Accurate modeling of sugarcane leaf diseases pose several challenges, including the need for large and diverse datasets, difficulty in differentiating between visually similar disease symptoms, and the adverse effects of environmental variability on model accuracy. Additionally, real-time prediction remains computationally intensive and often lacks generalizability across different crop types and geographical regions. To address these limitations, this paper proposes a novel framework—Wavelet Prompt-Tuned Multi-Task Taxonomic Transformer with Hierarchical Auto-Associative Polynomial Network (WATT-Net)—applied to the Sugarcane Leaf Image Dataset for effective disease prediction. Image pre-processing is enhanced through Discrete Wavelet Transformation combined with Pre-Gaussian Filtering (DWT-PGF) to reduce noise and blur, thereby improving image clarity. Region deviation analysis is employed to localize disease-affected areas, followed by Prompt-Tuned Multi-Task Taxonomic Transformer (PTMT)-based segmentation, which ensures precise boundary delineation. The new architecture (proposed PTMT architecture) does not manually engineer prompts but instead, they are learnt using a data-driven approach during training. The learned prompts represent task specific contextual priors and are dynamically adjusted to perform multi-tasks and their segmentation and classification better via formulating attention mechanism of the transformer The Scale-Aware Hierarchical Auto-Associative Polynomial Network (SA-HAAPNet) further strengthens the framework by extracting discriminative features for accurate classification. Disease region identification is refined using the Walk-Spread Algorithm (WSA), which contributes to higher detection accuracy and reduced error rates. Experimental results using Python-based implementation demonstrate superior performance, achieving 99.9% accuracy and 99.8% sensitivity, significantly outperforming existing models. The proposed WATT-Net approach offers a robust and scalable solution for real-time sugarcane leaf disease detection, with strong potential for broader agricultural applications.
No takes yet. Share an insight, caveat, or question.
Manjula et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: