Deep learning and IoT-based framework for sesame plant identification and weed detection

Area of Science:

• Agricultural Technology
• Computer Vision
• Machine Learning

Background:

• Automated weed identification is crucial for improving crop productivity and yield.
• Challenges in dense sesame fields include occlusion and varying lighting conditions, hindering accurate weed detection.
• Existing methods struggle with the complexities of real-world agricultural environments.

Purpose of the Study:

• To develop a hybrid deep learning model integrated with IoT for robust weed identification in sesame fields.
• To address challenges posed by occlusion and lighting variations in automated agricultural monitoring.
• To enhance crop productivity through precise weed detection and management.

Main Methods:

• A hybrid deep learning approach combining Region-based Convolutional Neural Network (R-CNN) and Support Vector Machine (SVM) was employed.
• Utilized approximately 1,300 YOLO-formatted annotated images for training and testing.
• The model incorporated CNN fine-tuning, CNN+SVM classification, and bounding box regression for object localization and identification.

Main Results:

• Achieved a classification accuracy of 99.73% for weed identification.
• Attained a maximum Intersection over Union (IoU) accuracy of 0.787 for bounding boxes.
• Demonstrated effective object detection and classification despite occlusion and lighting changes.

Conclusions:

• The proposed hybrid deep learning model offers a highly accurate solution for automated weed identification in challenging agricultural settings.
• The integration with IoT enhances the practical applicability of the system in real-time farming operations.
• This methodology shows significant potential for widespread adoption in agricultural sectors for efficient weed removal and yield enhancement.