LWheatNet: a lightweight convolutional neural network with mixed attention mechanism for wheat seed classification

Area of Science:

• Agricultural technology
• Computer vision
• Machine learning

Background:

• Deep learning in agriculture offers novel approaches to crop classification.
• Existing deep learning models face challenges like slow processing, high computational demands, and low accuracy for wheat seed image analysis, hindering real-time applications.

Purpose of the Study:

• To develop a lightweight wheat seed classification model that overcomes the limitations of existing deep learning approaches.
• To enhance feature representation and extraction for accurate wheat seed identification.

Main Methods:

• Proposed LWheatNet model integrating a mixed attention module with stacked inverted residual convolutional networks.
• Mixed attention mechanism combines parallel channel and spatial attention for improved feature representation.
• Stacked inverted residual networks utilize depthwise separable convolutions, channel shuffle, and channel split for efficient feature extraction, minimizing model size and computational load.

Main Results:

• LWheatNet achieved the highest performance among compared models (AlexNet, VGG16, MobileNet V2, MobileNet V3, ShuffleNet V2).
• Achieved 98.59% accuracy on the test set with a model size of only 1.33 M.
• Demonstrated superior performance compared to traditional Convolutional Neural Networks (CNNs) and other lightweight networks.

Conclusions:

• LWheatNet provides high recognition accuracy for wheat seed images while requiring minimal storage space.
• The model is suitable for real-time classification and recognition tasks on low-performance devices.