Automated Detection and Localization of Blastholes Through UAV Imaging and Machine Learning

Drilling accuracy is the primary factor influencing the explosive energy distribution of a blasting process. Therefore, knowing the final location of drillholes is crucial to obtain optimal blasting results. This research proposed a new methodology to control the drilling accuracy in open-pit mines using aerial images (captured through drones) and machine learning. The main focus of the study is to analyze if photogrammetry combined with machine learning can be used to detect drillholes in photogrammetry representations of blast patterns automatically. In this context, machine learning involves the development of algorithms trained to perform a task that implies a sort of human decision. To verify if machine learning can detect drillholes over photogrammetry representations, we started the project by collecting several datasets from different mine sites in Nevada (U.S.A.). Then, the images were processed to create photogrammetry representations of the drill patterns. These representations were Orthomosaic (union of photos) and Digital Elevation Models (DEM). In the process, thousands of patches were extracted and augmented from the photogrammetry representations. The patches were used to train and test SVM Models and CNN architectures to locate drillholes in cluttered scenes (typical in drill patterns). Once the training process reached enough validation accuracy, we tested the models with data left apart at the beginning of the process (testing datasets). The high recall, precision, and percentage of drillholes detected results obtained in our experiments demonstrate that using machine learning and photogrammetry is a feasible way to detect drillholes over photogrammetry representations. This novel technique provides a new solution for engineers to control the drilling accuracy, optimizing the blasting process by achieving more realistic simulations before implementing the designs.

Name thesis:degree_name

Doctor of Philosophy

Level thesis:degree_level

Doctorate Degree

Grantor

University of Nevada, Reno

Year dc:date.issued

2024

Author dc:creator

• Valencia, Jorge

Advisor dc:contributor.advisor

• Sattarvand, Javad

Committee members dc:contributor.committeemember

• Vahidi, Ehsan
• Chu, Pengbo
• Kallu, Raj
• Shoemaker, Kevin

• Convolutional Neural Networks
• Drill And Blast Engineering
• Machine Learning
• Mining Engineering
• Photogrammetry
• Support Vector Machine

Statement dc:rights

• Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International

Repository record dc:identifier.uri

https://scholarwolf.unr.edu/handle/11714/11050

OAI identifier oai:identifier

oai:scholarwolf.unr.edu:11714/11050