Deep learning-based temporal change detection of broadleaved weed infestation in rice fields using UAV multispectral...

Area of Science:

• Agricultural Science
• Remote Sensing
• Computer Science

Background:

• Optimizing herbicide application in rice cultivation is crucial for cost-efficiency and environmental sustainability.
• Current blanket spraying methods lead to excessive herbicide use and increased farming expenses.
• Site-specific weed management (SSWM) strategies require accurate, timely weed infestation data.

Purpose of the Study:

• To develop and evaluate a deep learning-based change detection method for monitoring broadleaved weed infestation dynamics in paddy fields.
• To assess the temporal changes in weed coverage using multispectral imagery from unmanned aerial vehicles (UAVs).
• To estimate potential herbicide savings achievable through targeted application strategies.

Main Methods:

• Collection of multispectral imagery using UAVs over rice fields.
• Development of a Deep Feedforward Neural Network (DFNN) for land cover classification (paddy, soil, broadleaved weeds).
• Application of post-classification comparison for temporal weed infestation assessment and change detection.

Main Results:

• Weed coverage consistently increased in untreated plots from 40.95% at 34 DAS to 47.43% at 48 DAS.
• Treated plots showed effective weed control, with potential herbicide savings estimated up to 40.95% at 34 DAS.
• A strong negative correlation (R²=0.9487) was observed between weed coverage and herbicide-saving potential.

Conclusions:

• Integrating UAV-based multispectral imaging with deep learning offers a powerful tool for temporal weed monitoring in rice cultivation.
• The developed approach facilitates precision agriculture by enabling data-driven decisions for herbicide application.
• This technology has the potential to significantly reduce herbicide usage and associated costs in rice farming.