LightWaveNet: a lightweight wavelet-enhanced high-low-frequency-aware network with multi-stage supervision for rice...

Area of Science:

• Agricultural Science
• Computer Vision
• Deep Learning

Background:

• Accurate rice disease identification is crucial for food security and intelligent agriculture.
• Existing deep learning models are computationally expensive and struggle to capture both fine-grained textures and structural features of diseased areas.

Purpose of the Study:

• To develop a lightweight and efficient deep learning network for accurate rice disease recognition.
• To address the limitations of existing models in handling high- and low-frequency information for comprehensive feature extraction.

Main Methods:

• Proposed LightWaveNet, a lightweight wavelet-enhanced high-low-frequency-aware network.
• Employed parallel wavelet convolution and max pooling for collaborative learning of frequency features.
• Utilized parallel max and average pooling during downsampling to preserve feature complementarity.
• Introduced multi-stage supervision for improved convergence and robustness.

Main Results:

• LightWaveNet achieved 95.90% recognition accuracy with only 0.28 M parameters and 0.02 G FLOPs.
• Demonstrated a favorable balance between accuracy and computational efficiency.
• Outperformed the Mobilenetv2 model in both accuracy and computational complexity.

Conclusions:

• LightWaveNet offers a feasible solution for rapid rice disease identification and intelligent prevention.
• Provides new insights into designing lightweight recognition networks for agricultural applications.
• Enables deployment on resource-constrained agricultural devices.