Open Set Self and Across Domain Adaptation for Tomato Disease Recognition With Deep Learning Techniques

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Deep learning systems offer eco-friendly plant disease monitoring.
• Current systems face performance degradation in new field conditions and with unseen data.

Purpose of the Study:

• To develop an open-set domain adaptation approach for robust plant disease recognition.
• To enable existing systems to operate effectively in new environments with unseen conditions and farms.

Main Methods:

• The proposed system treats plant disease diagnosis as an open-set learning problem.
• A framework with two deep learning modules performs bounding box detection and open-set domain adaptation.
• The detector enforces domain adaptation by classifying data as known classes or unknown.

Main Results:

• Extensive evaluation on a tomato plant disease dataset across three different farms was conducted.
• The approach demonstrated efficient adaptation to new field environments during field testing.
• Consistent performance gains were observed from explicitly modeling unseen data.

Conclusions:

• Open-set domain adaptation effectively addresses performance decay in plant disease recognition systems.
• The proposed method allows systems to maintain performance on known classes while handling unknown data in new environments.
• This research advances the reliability of automated plant disease monitoring in diverse field conditions.