Combining classifiers to detect faults in wastewater networks.

Area of Science:

• Civil Engineering
• Computer Science
• Machine Learning

Background:

• Sewer infrastructure requires regular inspection to identify structural faults.
• Current inspection methods can be labor-intensive and prone to human error.
• Automated analysis of Closed-Circuit Television (CCTV) footage offers a promising alternative.

Purpose of the Study:

• To develop and evaluate an enhanced methodology for automatic detection of structural faults in sewers using CCTV data.
• To compare the effectiveness of combining different machine learning classifiers for fault detection.
• To identify the optimal combination technique for improved accuracy and efficiency.

Main Methods:

• Utilized CCTV footage from real sewer surveys in the UK.
• Employed machine learning classifiers, specifically Support Vector Machine (SVM) and Random Forest (RF).
• Combined classifier outputs using three techniques: 'both', 'most likely', and 'stacking'.

Main Results:

• The 'stacking' technique demonstrated a 5% increase in detection accuracy compared to individual classifiers.
• All tested combination techniques showed minimal impact on processing efficiency.
• The stacking method proved to be the most effective for improving fault detection accuracy.

Conclusions:

• Combining machine learning classifiers, particularly using stacking, significantly enhances the accuracy of automatic sewer fault detection from CCTV footage.
• The developed methodology is suitable for practical implementation in sewer maintenance and management.
• This approach offers a more reliable and efficient alternative to traditional inspection methods.