SCANet: Implementation of Selective Context Adaptation Network in Smart Farming Applications

Area of Science:

• Computer Science
• Agricultural Science
• Artificial Intelligence

Background:

• Deep learning adoption in smart farming is uneven, with developing countries lagging.
• Smallholder farmers in developing nations have limited access to advanced agricultural technologies.
• Image classification is a key component of precision agriculture systems.

Purpose of the Study:

• To enhance image classification for smart farming applications.
• To address the limited adoption of deep learning in developing countries' agriculture.
• To leverage textural details of agricultural commodities for improved classification.

Main Methods:

• Proposed a deep learning approach named Selective Context Adaptation Network (SCANet).
• Implemented a feature enhancement strategy using level-wise information and context selection.
• Exploited contextual correlation features within crop images.

Main Results:

• SCANet achieved an accuracy of 88.72%.
• The proposed approach demonstrated the effectiveness of its context selection mechanism.
• Outperformed existing image classification methods in the domain.

Conclusions:

• SCANet offers a significant improvement in agricultural image classification.
• The model's effectiveness was validated on a real-world cocoa bean dataset from Indonesia.
• This work contributes to advancing precision agriculture technologies for wider adoption.