Deep Learning for Cadastral Boundary Extraction

Study Notes

Iterative Refinement

Analyze accuracy assessment results to identify discrepancies and areas of improvement
Refine pre-processing, image enhancement, segmentation, feature extraction, and boundary detection steps to optimize accuracy
Incorporate feedback from team members, domain experts, and stakeholders to refine the workflow and enhance the cadastral boundary extraction process

Deep Learning Algorithms

Convolutional Neural Networks (CNNs): Designed for analyzing visual data such as images
Generative Adversarial Networks (GANs): Involves two neural networks: a generator and a discriminator
Recurrent Neural Networks (RNNs): Well-suited for sequential data, including images with temporal dependencies

Fourier Transformation

Represents a one-dimensional function as a superposition of trigonometric sine and cosine terms
Calculates the frequency, amplitude, and phase of each sine wave needed to make up a signal
Assumes the signal is continuous with an infinite extent

Principal Component Analysis

Transforms a large set of variables into a smaller one, retaining most of the information
Used for multi-image manipulation, transforming coordinate axes of a multispectral dataset to a new set of mutually orthogonal coordinate axes

Control Points

Height Benchmarks: Used for vertical control, providing benchmark near works, and accurately leveled to National Benchmark
Classified into: Natural targets, Signalized targets, and Artificial points based on their appearance on imagery
Classified into: Control points, Check points, and Tie points based on their role in the adjustment

Machine Learning

Definition: A system capable of autonomous acquisition and integration of knowledge
Data Science Process: Data Collection, Data Preparation, EDA, Machine Learning, Visualization
Examples: Spam filtering, Credit card fraud detection, Detecting faces in images, MRI image analysis, Recommendation system

Bayesian Network Learning

Key Concepts: Nodes (attributes) = random variables, Conditional independence, Conditional probability table, Bayes Theorem
Settings: Known structure, fully observable (parameter learning), unknown structure, fully observable (structural learning), known structure, hidden variables (EM algorithm), unknown structure, hidden variables

Unsupervised Learning

Approaches: Clustering (similarity-based) and density estimation (e.g., EM algorithm)
Performance Tasks: Understanding and visualization, Anomaly detection, Information retrieval, Data compression

Lidar and Lidar-based Algorithms

Used for generating precise and directly geo-referenced spatial information
Lasers produce a coherent light source

Canny Edge Detector

Definition: A image processing technique used for detecting edges in digital images
Steps: Noise reduction, Gradient calculation, Non-maximum suppression, Double threshold, Edge Tracking by Hysteresis
Applications: Object Detection and Recognition, Image Segmentation, Quality Control and Inspection, Video processing, Image Retrieval, Robotics and Autonomous Vehicles

Support Vector Machine vs Box/Parallelepiped Classifiers

SVM: Supervised learning algorithm used for classification and regression tasks
Key idea: Maximize the margin between classes, and instances on the boundary of this margin are known as support vectors

Description

Iterate on pre-processing, image enhancement, segmentation, feature extraction, and boundary detection steps to optimize the accuracy of cadastral boundary extraction using deep learning techniques. Refine the workflow based on team feedback and domain expert insights.