Weed Classification from Natural Corn Field-Multi-Plant Images Based on Shallow and Deep Learning

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Automated weed control is crucial for modern agriculture but faces challenges in natural field conditions.
• Existing weed control methods are often limited to controlled environments, hindering real-world application.
• Accurate crop and weed identification is a prerequisite for developing effective automated weed management systems.

Purpose of the Study:

• To develop and evaluate a robust classification approach for distinguishing Zea mays (corn) from narrow-leaf weeds (NLW) and broadleaf weeds (BLW) using images captured in natural field conditions.
• To create and utilize a comprehensive image dataset of corn and weeds at various growth stages and locations.
• To compare the performance of a Convolutional Neural Network (CNN) based classification with a shallow learning approach for weed identification.

Main Methods:

• A large dataset of multi-plant images was collected under diverse natural field conditions.
• Regions of Interest (ROIs) were extracted using Connected Component Analysis (CCA).
• ROIs were classified using a CNN model and compared against a shallow learning method, with performance evaluated using accuracy, precision, recall, and F1-score.

Main Results:

• The CNN-based approach demonstrated superior performance in weed classification compared to the shallow learning method.
• The CNN model achieved a high accuracy of 97% for classifying corn, NLW, and BLW.
• The study confirmed the effectiveness of the CNN approach for weed classification in early growth stages within natural corn fields.

Conclusions:

• Convolutional Neural Networks offer a highly effective solution for crop and weed discrimination in complex, natural agricultural environments.
• The developed CNN-based system provides a promising foundation for advancing automated weed control technologies.
• Accurate plant classification using machine learning is vital for the future of precision agriculture and sustainable farming practices.