Enhancing the weed segmentation in diverse crop fields using computationally effective concatenated attention U-Net...

Area of Science:

• Agricultural Science
• Computer Vision
• Deep Learning

Background:

• Weeds significantly reduce crop yield by competing for resources.
• Uniform herbicide application is costly and environmentally concerning.
• Selective weed treatment requires advanced crop and weed segmentation systems.

Purpose of the Study:

• To develop a lightweight deep learning model for efficient crop and weed segmentation.
• To enable deployment of weed detection systems on edge devices for real-time application.
• To improve cost-effectiveness and environmental sustainability in agriculture.

Main Methods:

• Developed a novel convolutional neural network: Concatenated Attention U-Net with Convolutional Block Attention Module (CAUC).
• Integrated Linear Concatenated Blocks (LCB), Attention Gate (AG) connections, and Convolutional Block Attention Module (CBAM) for efficient feature utilization.
• Trained and validated the model using three datasets: Crop/Weed Field Image Dataset (CWFID), Sugar Beet, and Sunflower.

Main Results:

• Achieved high performance metrics: 99.09% Accuracy, 81.02% MIoU, and 99.06% F1-score.
• The CAUC model is lightweight, with a size of 5.6 MB and 0.377 million parameters.
• A functional computer vision application (13.7 MB) was created to demonstrate real-time efficacy on low-computational devices.

Conclusions:

• The CAUC model offers an effective and efficient solution for selective weed treatment.
• The lightweight design facilitates deployment on edge devices for practical agricultural applications.
• This approach presents a significant advancement in precision agriculture and sustainable farming practices.