A Novel Feature Selection Strategy Based on Salp Swarm Algorithm for Plant Disease Detection.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Deep learning excels in plant disease recognition but lacks interpretability.
• Handcrafted features offer personalized diagnosis but suffer from high dimensionality due to irrelevant features.

Purpose of the Study:

• To introduce a novel swarm intelligence algorithm, the Salp Swarm Algorithm for Feature Selection (SSAFS), for image-based plant disease detection.
• To optimize the selection of handcrafted features for improved classification accuracy and reduced dimensionality.

Main Methods:

• Developed and applied the Salp Swarm Algorithm for Feature Selection (SSAFS) to identify optimal handcrafted feature combinations.
• Compared SSAFS performance against five other metaheuristic algorithms.
• Evaluated methods using multiple metrics on UCI machine learning repository and PlantVillage phenomics datasets.

Main Results:

• SSAFS demonstrated superior performance in exploring feature spaces and identifying valuable features for diseased plant image classification.
• Experimental results and statistical analyses validated SSAFS's outstanding performance compared to state-of-the-art algorithms.
• The algorithm effectively maximized classification success while minimizing feature count.

Conclusions:

• SSAFS is a highly effective computational tool for feature selection in plant disease recognition.
• The proposed method significantly improves plant disease recognition accuracy and processing time.
• SSAFS offers a promising approach for personalized plant disease diagnosis in smart agriculture.