Related Experiment Videos
HASPNet: a hierarchically attentive signal-preserving network for papaya leaf disease classification with explainable
M Sundara Srivathsan1, Suchetha Manikandan1, S Preethi1
1Centre for Healthcare Advancements, Innovation and Research, Vellore Institute of Technology, Chennai, India.
Frontiers in Artificial Intelligence
|April 6, 2026
Summary
HASPNet accurately classifies papaya leaf diseases using a novel attention network, improving early disease detection in green farming. This method achieves high accuracy and efficiency, crucial for agricultural diagnostics.
Area of Science:
- Agricultural Science
- Computer Vision
- Plant Pathology
Background:
- Accurate papaya leaf disease classification is vital for early plant health surveillance and sustainable agriculture.
- Existing methods may lack the precision needed for fine-grained pathological analysis.
- The development of specialized deep learning models is essential for advancing green farming practices.
Purpose of the Study:
- To introduce HASPNet, a hierarchically attentive signal-preserving network for precise papaya leaf disease classification.
- To evaluate HASPNet's performance on the novel BDPapayaLeaf Dataset.
- To establish a new benchmark for papaya leaf disease detection systems.
Main Methods:
- Development of HASPNet, featuring a coordinated hierarchical attention framework with residual feature fusion and SE/CBAM modules.
- Optimization using Swish activation, depthwise separable convolutions, and a cosine warm-up learning rate schedule.
- Validation through ablation studies, comparison with state-of-the-art backbones, and Grad-CAM visualization for interpretability.
Main Results:
- HASPNet achieved 93.87% accuracy and a 94% F1-score on the BDPapayaLeaf Dataset.
- The model demonstrated superior performance and computational efficiency compared to MobileNetV2, DenseNet121, Inception-V3, Xception, and ResNet50.
- Swish activation was confirmed as the optimal non-linearity for this classification task, and inference time was reduced to 21.33 ms.
Conclusions:
- HASPNet offers a highly accurate and efficient solution for fine-grained papaya leaf disease classification.
- The model's interpretability and performance make it suitable for resource-constrained agricultural environments and real-world diagnostic systems.
- This work provides a valuable baseline for vision-based plant pathology and contributes to advancements in agricultural technology.
