Developing precision agriculture using data augmentation framework for automatic identification of castor insect...

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Castor (Ricinus communis L.) is a vital industrial crop, but its oil quality and yield are threatened by insect pests.
• Manual pest identification is time-consuming and requires expertise, hindering timely intervention.
• Automatic pest detection systems are crucial for precision agriculture and sustainable farming.

Purpose of the Study:

• To address the challenge of insufficient data for training effective vision-based insect pest detection models.
• To propose and evaluate a hybrid manipulation-based data augmentation approach for castor pests.
• To improve the accuracy and performance of deep convolutional neural networks for pest recognition.

Main Methods:

• A new hybrid manipulation-based data augmentation technique was developed.
• A dataset of common castor insect pests was created.
• Deep convolutional neural networks (VGG16, VGG19, ResNet50) were employed to analyze the augmentation method's impact.

Main Results:

• The proposed data augmentation method effectively addresses the issue of limited dataset size.
• The technique significantly improved the overall performance of pest detection models.
• The hybrid approach demonstrated superior results compared to previous methods.

Conclusions:

• The developed data augmentation strategy is effective for enhancing vision-based pest detection models.
• This approach supports the development of precision agriculture tools for castor cultivation.
• The findings contribute to more efficient and sustainable pest management strategies in agriculture.