Buscar
Filtrado por:
Departamento
Geography, Anthropology
Eliminar la restricciónDepartamento: Geography, Anthropology
Tema
Art, Geography, Anthropology, Archaeology, Computer Science
Eliminar la restricciónTema: Art, Geography, Anthropology, Archaeology, Computer Science
1 entrada encontrado
El número de resultados a mostrar por página
Resultados de la búsqueda
- Type:
- Dataset
- Descripción/Resumen:
- Q2000 Deep Learning Model Package This Technical Resource Bundle provides the Q200 Deep Learning model for open access download and use. The Q2000 DL model is built to detect Maya structures in Lidar data visualized at one meter per pixel. Currently, this repository contains the ESRI ArcGIS compatible DL model in the format .dlpk. We expect to convert the model in its current form into Pytorch (.pt) and TensorFlow (.h5) formats to incude them also here for user access. The Q2000.dlpk file was created in 2025 at University of Cincinnati by Benjamin Britton, using ArcGIS Pro v.3.3 with data from the NASA Ames G-LiHT. It is intended as an experiment to evaluate the practicability of creating a broadscale deep learning model that can be used effectively to identify Maya structures in Lidar data across the length and breadth of the Yucatan Peninsula. The Q2000 model is the subject of an article, a draft of which is also included in this dataset, called "Evaluating Broadscale Deep Learning for Maya Settlement Detection in G-LiHT Lidar" which examines the process and rationale of the model development in detail. The article has been accepted for publication and this site is included in that article with a link to this permanent (DOI) publication site. To use the model with ArcGIS Pro, use a Lidar dataset converted to a 1m/pixel DEM file and visualized as a 3-channel RGB Hillshade or other customized visualization as source input. -Using the ArcGIS Pro Spatial Analyst Extension, the geoprocessing tool called "Detect Object Using Deep Learning" may be invoked. -For Input Raster, add your Lidar visualization (a Hillshade visualization might be easiest for most users). -For Output Detected Objects, specify a new Layer name, and this will be the layer on which the detections will be recorded and displayed. -For Model Definition, use Q2000.dlpk. -Unless you want to run "arguments", you can leave the Arguments boxes as their Default. -I suggest checking the box (On) for Non Max suppression because it will reduce the amount of overlapping detections if target objects are located very close to each other, and I suggest a Non-Maximum Suppression (NMS) ratio of 0.5. This will tend to suppress detections that overlap by more than 50 percent. -I suggest you use Pixel Space unchecked (Off), since it is for an unrelated sort of object detection. -Before you click run, open the "Environments" tab (at the top of the window, next to the "Parameters" tab). Leave all the settings at their defaults there - except scroll down to the bottom of Parameters tab to the section called "Processor Type", pull down the Processor type pull-down, and choose GPU (it is set to CPU by default). -Then click Run and it will generate a new layer showing its detections as bounding boxes around target objects. You can see details for each detection by opening the Attribute Table on the new layer. You can see a screen capture of such a configuration in the image called Q200DemoScreenCap.jpg, included in this site's dataset, showing a detection on G-LiHT transect Yucatan_South_GLAS_395 near Pixoyal, with a detection of a Maya staircase highlighted on the display, and its corresponding information highlighted in the Attribute Table for it.
- Creador/Autor:
- Britton, Benjamin
- Peticionario:
- Benjamin Britton
- Fecha modificada:
- 08/23/2025
- Fecha modificada:
- 11/13/2025
- Fecha de creacion:
- March 15, 2025
- Licencia:
- Public Domain Mark 1.0