Multi-Class Weed Recognition Using Hybrid CNN-SVM Classifier.

Area of Science:

• Agricultural technology
• Computer vision
• Machine learning

Background:

• Deep learning models like Convolutional Neural Networks (CNNs) are crucial for analyzing field conditions in agriculture.
• Accurate weed identification is essential for optimizing crop yields and reducing herbicide use.

Purpose of the Study:

• To develop and evaluate novel CNN-SVM models for automated weed classification.
• To enhance the accuracy of weed recognition in agricultural settings using deep learning.

Main Methods:

• Feature extraction using pre-trained CNN models (ResNet-50 and VGG16).
• Classification of extracted features using Support Vector Machines (SVM).
• Training and testing on the public DeepWeeds multi-class dataset.

Main Results:

• The proposed ResNet-50-SVM model achieved 97.6% accuracy, and the VGG16-SVM model achieved 95.9% accuracy on the DeepWeeds dataset.
• These results represent an improvement of 1.5% and 2.7% over state-of-the-art methods, respectively.
• The models demonstrated high recognition accuracy for various weed species.

Conclusions:

• The developed ResNet-50-SVM and VGG16-SVM approaches are effective for autonomous weed classification in agriculture.
• These CNN-SVM models offer a promising solution for boosting farming productivity through precise field condition inference.
• The study highlights the potential of integrating CNNs with SVMs for advanced agricultural monitoring systems.