Scholarlink

Abstract <p>Determining the spatial resolution (GSD) of remote sensing images is in demand in various fields, from environmental monitoring to urban planning and agriculture, which makes it relevant to analyze both urban and natural landscapes. This study focuses on deep learning methods for evaluating the GSD of still images, and its purpose is to study the impact exerted on the quality of GSD assessment by approaches, such as autoencoder pre-training on still images (SSL) and taking into account features from different levels of the model. All the models under consideration were trained on images of landscapes of different types. Scaling augmentations were also used to train them to expand the range of GSD in data samples. ResNet18 was used as a baseline model, which, based on available data, demonstrates MAPE values ranging from 2.73 to 15.6%. Using the relative loss function directly in training allows one to improve performance on data with low spatial resolution from 15.6 to 14.7%. At the same time, using the SSL approach slightly improves performance at high GSD values. A combination of SSL with FPN reaches 3.61% on a set with a mixed landscape type. The results make it possible to use the trained models to solve applied problems. However, it is necessary to choose the most appropriate method for each specific case.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Neural Network Assessment of Spatial Resolution of Multi-Scale Aerial Photographs of Urban and Natural Landscapes

  • 期刊论文

  • 发表日期：

    2025年8月1日

查看全文
  • A. V. Zabolotskiy,
  • K. V. Sobyanin

摘要

Abstract

Determining the spatial resolution (GSD) of remote sensing images is in demand in various fields, from environmental monitoring to urban planning and agriculture, which makes it relevant to analyze both urban and natural landscapes. This study focuses on deep learning methods for evaluating the GSD of still images, and its purpose is to study the impact exerted on the quality of GSD assessment by approaches, such as autoencoder pre-training on still images (SSL) and taking into account features from different levels of the model. All the models under consideration were trained on images of landscapes of different types. Scaling augmentations were also used to train them to expand the range of GSD in data samples. ResNet18 was used as a baseline model, which, based on available data, demonstrates MAPE values ranging from 2.73 to 15.6%. Using the relative loss function directly in training allows one to improve performance on data with low spatial resolution from 15.6 to 14.7%. At the same time, using the SSL approach slightly improves performance at high GSD values. A combination of SSL with FPN reaches 3.61% on a set with a mixed landscape type. The results make it possible to use the trained models to solve applied problems. However, it is necessary to choose the most appropriate method for each specific case.