An optimized detection model for micro-terrain around transmission lines.

Area of Science:

• Geoinformatics and Remote Sensing
• Electrical Engineering and Power Systems

Background:

• Accurate micro-terrain detection is crucial for transmission line layout and maintenance.
• Traditional methods suffer from incompleteness, classification ambiguity, and inefficiency, especially in distinguishing complex terrain like saddles and canyons.

Purpose of the Study:

• To optimize micro-terrain feature calculation and detection strategies for transmission line environments.
• To develop and evaluate a novel GPU parallel random forest method for efficient and accurate micro-terrain detection.

Main Methods:

• Utilized a GPU parallel random forest model for micro-terrain feature extraction.
• Applied the method to Digital Elevation Model (DEM) data from 49 representative transmission line routes in complex mountainous terrain.
• Optimized feature calculation and detection strategies to address limitations of traditional approaches.

Main Results:

• Achieved a classification accuracy of 97.96% and a Kappa coefficient of 0.974 for identifying typical (saddle, canyon, alpine watershed, uplift) and atypical micro-terrain types.
• Demonstrated significant performance improvement over traditional methods (75.19% accuracy, 0.642 Kappa).
• Attained substantial efficiency gains with training and classification acceleration ratios of 50.57 and 109.06, respectively, due to GPU parallelization.

Conclusions:

• The proposed GPU parallel random forest method offers a highly accurate and efficient solution for micro-terrain detection around transmission lines.
• This advancement provides more reliable micro-terrain data, crucial for enhancing transmission line maintenance, layout planning, and disaster risk mitigation.