Automated Rock Classification with depth analysis for Borehole Logging Using YOLOProject OverviewIn geological surveys, a substantial number of boreholes are drilled and analyzed to create 3D models of the underground. Typically, this involves classifying rock sections from each borehole core, a time-consuming task for geologists. Each borehole can be 500 to 2500 meters deep, and with 50 to 100 boreholes, the manual classification process can take months. To address this, we developed an object detection model using YOLO (You Only Look Once) to automate rock classification in borehole cores.ChallengesVolume of Data: Each borehole core is several kilometers long.Manual Classification: A skilled geologist can classify only 50 to 70 meters per day.High Costs: Geologists' salaries are substantial, and the manual process is resource-intensive.SolutionUsing a Convolutional Neural Network (CNN) and the YOLO computer vision model, we automated the rock classification process. YOLO is well-suited for this task due to its real-time object detection capabilities and high accuracy.Key AchievementsEfficiency: The model can classify borehole cores in minutes, compared to months for manual classification.Cost Reduction: By automating the classification, significant savings on geologists' salaries are achieved.WorkflowData Collection: High-resolution images of borehole cores were collected and labeled by expert geologists to create a training dataset.Model Training: The YOLO model was trained on this dataset to detect and classify different rock types.Model Deployment: The trained model was deployed to process new borehole core images, providing classifications in real-time.Validation and Testing: The model's performance was validated against a test set and further refined based on feedback.ImpactTime-Saving: Reduced the classification time from months to minutes.Economic Efficiency: Lowered the overall cost of geological surveys.

Generative AI Internship | Think Turbo Jul '24 - Aug '24• Developed text-summarization model leveraging Google's Pegasus model fine-tuned on the CNN/DailyMail dataset from Hugging Face• Implemented a professional project setup utilizing modular coding and OOPs principles, enhancing code maintainability and scalability• Created the Docker environment, and deployed the model on AWS, decreasing the model latency up to sub-500 ms leveraging modular coding