Superpixel-Based Shallow Convolutional Neural Network (SSCNN) for Scanned Topographic Map Segmentation

Abstract

Motivated by applications in topographic map information extraction, our goal was to discover a practical method for scanned topographic map (STM) segmentation. We present an advanced guided watershed transform (AGWT) to generate superpixels on STM. AGWT utilizes the information from both linear and area elements to modify detected boundary maps and sequentially achieve superpixels based on the watershed transform. With achieving an average of 0.06 on under-segmentation error, 0.96 on boundary recall, and 0.95 on boundary precision, it has been proven to have strong ability in boundary adherence, with fewer over-segmentation issues. Based on AGWT, a benchmark for STM segmentation based on superpixels and a shallow convolutional neural network (SCNN), termed SSCNN, is proposed. There are several notable ideas behind the proposed approach. Superpixels are employed to overcome the false color and color aliasing problems that exist in STMs. The unification method of random selection facilitates sufficient training data with little manual labeling while keeping the potential color information of each geographic element. Moreover, with the small number of parameters, SCNN can accurately and efficiently classify those unified pixel sequences. The experiments show that SSCNN achieves an overall F1 score of 0.73 on our STM testing dataset. They also show the quality of the segmentation results and the short run time of this approach, which makes it applicable to full-size maps.

Combining Remote-Sensing-Derived Data and Historical Maps for Long-Term Back-Casting of Urban Extents

“…Future work will include the test of alternative clustering or unsupervised classification techniques that infer the optimum number of clusters from the data [ 94 , 95 ], as well as the use of different color space representations [ 96 ]. Future work will also include the performance of object-based methods [ 57 , 64 ] as compared to the pixel-based extraction method presented herein.…”

“…Such recent efforts include the mining of (historical) map collections by their content or associated metadata [ 32 - 37 ], automated georeferencing [ 18 , 38 - 40 ] and alignment [ 41 , 42 ], text detection and recognition [ 43 - 45 ], and the extraction of thematic map content, often involving (deep) machine learning methods, focusing on specific geographic features such as forest [ 46 ], railroads [ 33 , 47 ], road network intersections [ 48 , 49 ] and road types [ 50 ], archeological content [ 51 ] and mining features [ 52 ], cadastral parcels boundaries [ 53 , 54 ], wetlands and other hydrographic features [ 55 , 56 ], linear features in general [ 57 ], land cover/land use [ 58 ], urban street networks and city blocks [ 34 ], building footprints [ 13 , 59 , 60 ], and historical human settlement patterns [ 61 - 63 ]. Other approaches use deep-learning-based computer vision for generic segmentation of historical maps [ 64 , 65 ], generative machine learning approaches for map style transfer [ 66 , 67 ], or attempt to mimic historical overhead imagery based on historical maps [ 68 ].…”

Semantic Segmentation Deep Learning for Extracting Surface Mine Extents from Historic Topographic Maps

“…Li et al [24] explored the recognition of text on topographic maps using DL while Uhl et al [25] investigated the mapping of human settlements from maps using CNNs. Lui et al [41] explored CNNs for the general segmentation of topographic maps. There has been some success in applying DL methods to other types of digital data, beyond the primary focus on multispectral satellite or aerial imagery.…”

Combining Remote Sensing Derived Data and Historical Maps for Long-Term Back-Casting of Urban Extents

“…Such recent efforts include the mining of (historical) map collections by their content or associated metadata [32][33][34][35][36][37], automated georeferencing [18,[38][39][40] and alignment [41,42], text detection and recognition [43][44][45], or the extraction of thematic map content, often involving (deep) machine learning methods, focusing on specific geographic features such as forest [46], railroads [33,47], road network intersections [48,49] and road types [50], archeological content [51] and mining features [52], cadastral parcels boundaries [53,54], wetlands and other hydrographic features [55,56], linear features in general [57], land cover / land use [58], urban street networks and city blocks [34], building footprints [13,59,60] and historical human settlement patterns [61][62][63]. Other approaches use deep learning based computer vision for generic segmentation of historical maps [64,65], generative machine learning approaches for map style transfer [66,67] or attempt to mimic historical overhead imagery based on historical maps [68].…”