Application of Multilayer Perceptron with Automatic Relevance Determination on Weed Mapping Using UAV Multispectral...

Area of Science:

• Agricultural Science
• Remote Sensing
• Machine Learning

Background:

• Remote sensing is crucial for plant species discrimination and weed mapping.
• Accurate identification of invasive or problematic plant species is essential for effective land management.

Purpose of the Study:

• To demonstrate the successful detection and mapping of Silybum marianum using multilayer neural networks.
• To evaluate the accuracy of the Multilayer Perceptron with Automatic Relevance Determination (MLP-ARD) for Silybum marianum identification.

Main Methods:

• Utilized a fixed-wing unmanned aerial vehicle (UAV) equipped with a multispectral camera (green, red, near-infrared).
• Acquired high-resolution (0.1 m) aerial imagery.
• Employed the Multilayer Perceptron with Automatic Relevance Determination (MLP-ARD) model, incorporating spectral bands and texture features as input.

Main Results:

• Achieved a high identification accuracy rate of 99.54% for Silybum marianum.
• Successfully differentiated Silybum marianum from other vegetation, primarily Avena sterilis L.
• Demonstrated the effectiveness of MLP-ARD in analyzing multispectral UAV imagery.

Conclusions:

• The study highlights the significant potential of MLP-ARD for accurate Silybum marianum mapping using multispectral UAV data.
• The high accuracy achieved suggests this method is a promising tool for agricultural and ecological monitoring.
• Further research could explore the application of this technique over longer durations and diverse environments.