Segmentation of Satellite Imagery using U-Net Models for Land Cover Classification

Abstract

The focus of this paper is using a convolutional machine learning model with a modified U-Net structure for creating land cover classification mapping based on satellite imagery. The aim of the research is to train and test convolutional models for automatic land cover mapping and to assess their usability in increasing land cover mapping accuracy and change detection. To solve these tasks, authors prepared a dataset and trained machine learning models for land cover classification and semantic segmentation from satellite images. The results were analysed on three different land classification levels. BigEarthNet satellite image archive was selected for the research as one of two main datasets. This novel and recent dataset was published in 2019 and includes Sentinel-2 satellite photos from 10 European countries made in 2017 and 2018. As a second dataset the authors composed an original set containing a Sentinel-2 image and a CORINE land cover map of Estonia. The developed classification model shows a high overall F 1 score of 0.749 on multiclass land cover classification with 43 possible image labels. The model also highlights noisy data in the BigEarthNet dataset, where images seem to have incorrect labels. The segmentation models offer a solution for generating automatic land cover mappings based on Sentinel-2 satellite images and show a high IoU score for land cover classes such as forests, inland waters and arable land. The models show a capability of increasing the accuracy of existing land classification maps and in land cover change detection.

Comparing CNNs and Random Forests for Landsat Image Segmentation Trained on a Large Proxy Land Cover Dataset

“…Although these results appear to not completely meet well-known benchmarks for land cover classifications such as 85% OA and 70% per-class accuracy (e.g., [59]), they were broadly in line with similar studies that use label data derived from land use or land cover mapping in combination with Landsat or Sentinel-2 inputs (e.g., [49,50,[60][61][62]).…”

“…For the U-Net CNNs, transfer learning [43] was employed using various ResNet models [48] as the encoder portion of the U-Net model, initialised with weights obtained from training using the ImageNet ILSVRC-2012-CLS dataset [42]. Several remote sensing studies have used ImageNet-trained ResNet encoder weights with the U-Net models for segmentation to produce land cover maps, e.g., [49][50][51]. We compared the results from ResNet encoder versions with varying numbers of neural network layers (ResNet18, ResNet34, ResNet50, ResNet101, ResNet152).…”

L^3U-net: Low-Latency Lightweight U-net Based Image Segmentation Model for Parallel CNN Processors

“…These architectures are different from [26] as there are residual connections between specified encoding and decoding layers. Various modified U-net models have also been used for many different applications like 3D image segmentation [9] or land use determination [8,32,35]. U-net is still a prevalent approach, and many recent studies propose modifications like adding dense skip connections [14] or proposing attention-based U-net architectures [12,16] for different applications.…”

Edge Tracing using Gaussian Process Regression

“…Semantic image segmentation is an important tool for deriving information from an image and plays a crucial role in many applications, facilitating higher-level image analysis [1,2,3]. The goal of any image segmentation algorithm is to partition an image into meaningful sub-regions.…”