MAVM-UNet: multiscale aggregated vision MambaU-Net for field rice pest detection

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Rice production is threatened by pests, leading to yield loss and environmental issues from pesticide overuse.
• Accurate detection of diverse rice pests is challenging due to their irregular shapes, small size, and complex backgrounds.
• Effective pest detection is crucial for precise pest management and sustainable agriculture.

Purpose of the Study:

• To develop an advanced deep learning model for accurate and efficient detection of rice field pests.
• To address the challenges posed by pest variability and complex field conditions in image-based detection.
• To provide a foundation for improved pest monitoring and control strategies in rice cultivation.

Main Methods:

• A novel multiscale aggregated vision MambaU-Net (MAVM-UNet) model was constructed.
• The model incorporates Visual State Space (VSS), multiscale VSS (MSVSS), Channel-Aware VSS (CAVSS), and multiscale attention aggregation (MSAA) modules.
• CAVSS and MSAA modules were integrated to capture multi-scale features, low-level details, and high-level semantics.

Main Results:

• The MAVM-UNet model achieved superior performance compared to state-of-the-art models on the IP102 dataset.
• The model obtained a Precision (PA) of 82.07% and Mean Intersection over Union (MIoU) of 81.48%.
• Experimental results demonstrate the model's effectiveness in capturing both fine-grained and coarse-grained pest features.

Conclusions:

• The developed MAVM-UNet model offers a significant advancement in automated rice pest detection.
• This technology can guide more precise and environmentally friendly pest monitoring and control in rice fields.
• The study provides valuable insights for applying advanced computer vision techniques in agricultural pest management.