An improved CNN model in image classification application on water turbidity.

Area of Science:

• Environmental Science
• Computer Science
• Artificial Intelligence

Background:

• Water turbidity is a key indicator of water clarity, crucial for environmental protection and ecological balance.
• Classifying subtle differences in water turbidity images presents a significant challenge due to the fine-grained nature of the changes.
• Convolutional Neural Networks (CNNs) are powerful tools for image classification and feature extraction.

Purpose of the Study:

• To explore the application of CNNs for classifying water turbidity from images.
• To optimize CNN models for improved prediction accuracy and efficiency in water turbidity classification.
• To investigate the effectiveness of different CNN architectures in handling subtle image variations.

Main Methods:

• Proposed four distinct CNN models for water turbidity classification.
• Adjusted the number of layers within the CNN models to enhance prediction accuracy.
• Conducted experiments on both noise-free and noisy datasets to assess model performance.
• Evaluated models based on classification accuracy and processing time.

Main Results:

• The CNN-10 model, incorporating a dropout layer, achieved a classification accuracy of 96.5% on noisy datasets.
• Experiments demonstrated varying performance across the four proposed CNN models.
• The study highlighted the impact of model architecture, specifically layer adjustments, on prediction accuracy.

Conclusions:

• CNNs are effective for fine-grained image classification tasks, specifically in water turbidity assessment.
• The optimized CNN-10 model shows high potential for accurate and efficient water turbidity classification.
• This research opens new avenues for applying deep learning in environmental monitoring and water quality analysis.