HAMNet: Hierarchical Multi-scale Attention Network for precise disease detection in pearl millet using spatial fusion

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Pearl millet is crucial for arid/semi-arid agriculture but faces significant yield losses from fungal and bacterial diseases.
• Traditional disease detection methods are inefficient, subjective, and time-consuming.
• Existing deep learning models struggle with accurate disease localization and segmentation precision.

Purpose of the Study:

• To develop an advanced AI framework for precise and automated detection of pearl millet diseases.
• To enhance disease localization, segmentation accuracy, and classification performance in high-resolution imagery.

Main Methods:

• Proposed HAMNet-Mask R-CNN: a hierarchical multi-scale attention network with spectral-spatial fusion.
• Utilized hierarchical multi-scale feature learning and an attention mechanism for critical region prioritization.
• Employed Mask R-CNN for pixel-wise segmentation and boundary refinement.
• Implemented using Python and TensorFlow on high-resolution image datasets.

Main Results:

• HAMNet-Mask R-CNN achieved superior performance: 99.35% Dice score, 98.80% IoU, 99.75% Precision, 99.78% Recall, and 99.65% Accuracy.
• Outperformed U-Net, SegNet, DeepLabV3, and FCN in disease segmentation and classification metrics.
• Demonstrated enhanced accuracy in localizing and segmenting disease-affected areas.

Conclusions:

• HAMNet-Mask R-CNN offers a robust and scalable solution for real-time disease monitoring in smart agriculture.
• The integration of hierarchical attention and spectral-spatial fusion significantly boosts detection accuracy.
• The model contributes to improved agricultural productivity through reliable disease identification.