A new pairwise deep learning feature for environmental microorganism image analysis.

Area of Science:

• Environmental microbiology
• Machine learning
• Bioinformatics

Background:

• Environmental microorganisms (EMs) are vital for pollution control, sanitation, and pollutant decomposition.
• Accurate identification of suitable EMs is essential for effective environmental remediation.
• Current identification methods can be slow, costly, and lack consistency.

Purpose of the Study:

• To develop a novel method for rapid, cost-effective, and accurate identification of environmental microorganisms.
• To introduce pairwise deep learning features (PDLFs) for enhanced microorganism analysis.
• To improve the consistency and accuracy of EM classification.

Main Methods:

• Leveraged Shi and Tomasi interest points for deep learning feature extraction from image patches.
• Applied Delaunay triangulation and straight line geometric theorems to pair deep learning features.
• Integrated handcrafted and deep learning features using the PDLFs technique.
• Classified EMs using Support Vector Machines (SVMs), Linear Discriminant Analysis (LDA), Logistic Regression, XGBoost, and Random Forest classifiers.

Main Results:

• The PDLFs technique achieved outstanding classification results, with accuracies up to 99.56%.
• Significant improvements were observed compared to non-paired deep learning features: accuracy increased by 5.95%, F1-score by 62.40%, recall by 62.37%, precision by 61.84%, and specificity by 3.23% across various classifiers.
• The proposed method demonstrated superior performance in classifying environmental microorganisms.

Conclusions:

• Pairwise deep learning features (PDLFs) offer a highly effective approach for environmental microorganism identification.
• This novel technique enhances classification accuracy, efficiency, and consistency.
• PDLFs represent a significant advancement in applying machine learning to environmental microbiology challenges.