Environmental microorganism classification using optimized deep learning model.

Area of Science:

• Environmental microbiology
• Artificial intelligence
• Computer vision

Background:

• Accurate identification of environmental microorganisms (EM) is crucial for assessing water quality.
• Deep convolutional neural networks (CNNs) are powerful tools for image classification tasks.

Purpose of the Study:

• To evaluate and optimize CNN models for classifying EM from microscopic images.
• To enhance the accuracy of EM classification using techniques like data augmentation and genetic algorithm optimization.

Main Methods:

• Transfer learning was applied to three popular CNNs: ResNet50, Vgg16, and Inception-v3.
• The Environmental Microorganism Dataset (EMDS), containing 294 images across 21 classes, was used for training and testing.
• Data augmentation and optimization of the fully connected layer (neuron number and dropout rate) using a genetic algorithm (GA) were employed.

Main Results:

• Inception-v3 achieved 84.9% accuracy, outperforming ResNet50 and Vgg16.
• Data augmentation improved Inception-v3 performance.
• Optimized Inception-v3 with GA reached 90.5% accuracy.
• The optimized Inception-v3 model combined with data augmentation achieved 92.9% accuracy, a 21% improvement over Vgg16.

Conclusions:

• Optimized Inception-v3 with data augmentation offers a highly accurate solution for EM classification.
• This AI-driven approach can be integrated with microscopy and digital camera systems for effective environmental monitoring.
• The optimized model potentially requires fewer neurons compared to Vgg16, suggesting computational efficiency.