Insulator detection in transmission line based on Log AdaBoost

Area of Science:

• Electrical Engineering
• Computer Vision
• Machine Learning

Background:

• Smart grid operation relies on accurate insulator detection.
• Traditional image processing methods struggle with background interference in insulator images.

Purpose of the Study:

• To propose a new Log AdaBoost method for robust insulator detection.
• To improve detection accuracy and generalization in smart grids.

Main Methods:

• Developed a boosting algorithm optimizing Polylog loss using gradient descent.
• Implemented a new weight updating strategy for optimal weak classifier selection.
• Introduced a neighborhood Haar-like feature to enhance insulator-background contrast.

Main Results:

• Achieved the lowest test error on 11 of 20 UCI datasets.
• Demonstrated lower testing error and smaller margin variance on the ACDC dataset.
• Attained an AUC of 0.82 on the CPLID insulator detection dataset with 21k parameters.

Conclusions:

• The proposed Log AdaBoost method offers superior performance for insulator detection.
• The algorithm shows improved accuracy, generalization, and efficiency over existing AdaBoost variants.
• This method is effective for real-world smart grid applications.