RICE WEED DETECTION AND CLASSIFICATION USING UAV

Abstract

Weed detection in rice crops is essential for improving rice yield, as weeds compete with the plants for nutrients, water, and sunlight. Timely identification and control of weeds can significantly enhance rice growth. Traditional methods of weed detection are labor-intensive and inefficient, especially for large fields. This project proposes a novel approach for rice weed detection using UAV-captured images and the deep learning model GoogLeNet. The UAVs capture high-resolution RGB images of rice fields at various growth stages, which are then used to train the GoogLeNet model. The model, implemented in MATLAB , accurately detects and classifies weeds in rice crops. The results show that GoogLeNet offers an efficient and effective solution for automated weed detection, enabling targeted weed control and contributing to increased rice production The use of UAV imagery combined with GoogLeNet not only streamlines the weed detection process but also enhances accuracy by leveraging the model’s deep feature extraction capabilities. The system’s performance was evaluated using precision, recall, and F1-score metrics, demonstrating high classification accuracy across diverse field conditions. This approach minimizes the need for blanket herbicide application, promoting environmentally sustainable farming practices. Moreover, the adaptability of the system allows it to be integrated into existing precision agriculture frameworks, providing farmers with actionable insights for site-specific weed management. Future work may involve expanding the dataset to include various rice varieties and weed species, optimizing the model for real-time processing, and exploring multi-spectral or hyperspectral image integration to further improve detection accuracy. This research highlights the potential of combining UAV technology with deep learning for scalable, intelligent weed control solutions in modern rice farming..