A Combinatorial Solution to Point Symbol Recognition

Abstract

Recent work has shown that recognizing point symbols is an essential task in the field of map digitization. For the identification of symbols, it is generally necessary to compare the symbols with a specific criterion and find the most similar one with each known symbol one by one. Most of the works can only identify a single symbol, a small number of works are to deal with multiple symbols simultaneously with a low recognition accuracy. Given the two deficiencies, this paper proposes a deep transfer learning architecture, where the task is to learn a symbol classifier with AlexNet. For the insufficient dataset, we develop a method for transfer learning that uses a MNIST dataset to pretrain the model, which makes up for the problem of small training dataset and enhances the generalization of the model. Before the recognition process, preprocessing the point symbols in the map to coarse screening out the areas suspected of point symbols. We show a significant improvement over using point symbol images to keep a high performance in being able to deal with many more categories of symbols simultaneously.

An automation solution to convert CAD engineering drawings into railroad station models

“…Statistical methods utilize statistical or machine learning techniques to generate features according to values and distributions of pixels of images and then calculate the similarities between patterns and target symbols (Pham et al, 2015;Yang et al, 2008) or train learning-based models to predict the probabilities of matching target symbols. For instance, Li et al (2019) and Fu and Kara (2011) utilized neural networks to learn symbol features and Quan et al (2018) proposed a combinatorial learning solution for point symbol recognition. Since the accuracy of statistical methods is significantly impacted by the way of creating features, many workers studied various feature engineering methods, such as kernel density (W. Zhang et al, 2006), sparse representation (Do et al, 2016), and improved generalized Hough transform (J.…”

“…(2019) and Fu and Kara (2011) utilized neural networks to learn symbol features and Quan et al. (2018) proposed a combinatorial learning solution for point symbol recognition. Since the accuracy of statistical methods is significantly impacted by the way of creating features, many workers studied various feature engineering methods, such as kernel density (W. Zhang et al., 2006), sparse representation (Do et al., 2016), and improved generalized Hough transform (J.…”

Automatic vectorization of point symbols on archive maps using deep convolutional neural network

“…Saeedimoghaddam and Stepinski used CNN to detect road intersection points and achieved an average of 90% accuracy, with 82% of intersection points extracted (Saeedimoghaddam and Stepinski 2020). Quan et al focused on point symbol recognition combined with image segmentation on a pretrained CNN of AlexNet architecture and achieved 98.97% accuracy on single test images (Quan et al 2018). In this study, an improved CNN was used to detect four distinct point symbols simultaneously on scanned historical topographic maps.…”

Machine Recognition of Map Point Symbols Based on YOLOv3 and Automatic Configuration Associated with POI

“…By comparing the template image with the test image, the similarity between the two is calculated to locate the predefined target. In addition, there is a statistical analysis [8,9] method for characterizing texture images with model coefficients, a structural analysis [10,11] method for recognizing complex objects based on image structural features, a mathematical morphological method [12][13][14] for studying image spatial shape and structure using set theory, and a deep learning method [15][16][17] for machines to simulate the operation of human brain thinking, etc. Traditional Convolutional Neural Networks (hereinafter referred to as CNN) and Deep Neural Networks (DNN) [18][19][20] struggle to comprehend the complex semantic information of symbols, whereas the development of deep learning [21] in the field of graphics is gradually resolving traditional graphics problems.…”