Lightweight and efficient neural network with SPSA attention for wheat ear detection.

Area of Science:

• Computer Vision
• Agricultural Technology
• Deep Learning

Background:

• Deep neural networks have advanced crop detection.
• Wheat ear detection is challenging due to complex backgrounds, dense targets, and overlaps.
• Existing detectors often struggle to balance computational cost and precision for real-world deployment.

Purpose of the Study:

• To develop a lightweight and efficient wheat ear detector.
• To improve the accuracy and speed of wheat ear detection in complex agricultural environments.
• To address the limitations of current detectors in balancing performance and computational resources.

Main Methods:

• Utilized a lightweight backbone network with asymmetric convolution for feature extraction.
• Introduced Shuffle Polarized Self-Attention (SPSA) to adaptively select features across spatial and channel dimensions.
• Employed TanhExp activation function for faster inference and CIOU loss for improved border regression, especially for occluded targets.

Main Results:

• The proposed SPSA detector demonstrated superior detection performance on the Global Wheat Head Detection dataset.
• Achieved a better balance between computational cost and detection precision compared to state-of-the-art methods.
• Effectively handled challenges like dense targets and overlapping wheat ears.

Conclusions:

• The SPSA wheat ear detector offers an efficient and accurate solution for agricultural applications.
• The method provides a practical approach for real-world deployment of wheat ear detection systems.
• The integration of SPSA, TanhExp, and CIOU loss significantly enhances detection capabilities.