Semantic Segmentation of Remote Sensing Imagery Using Object-Based Markov Random Field Model With Regional Penalties

Chen, Zheng; Wang, Leiguang

doi:10.1109/jstars.2014.2361756

Cited by 57 publications

(22 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Previously in remote sensing only a single classification hierarchy (either LC or LU) was modelled and predicted, such as via the Markov Random Field with Gibbs joint distribution for LC characterisation (e.g. Schindler, 2012;Zheng and Wang, 2015;Hedhli et al, 2016). They are essentially designed to fit a model that can link the land cover labels x to the observations y (e.g.…”

Section: Joint Deep Learning Modelmentioning

confidence: 99%

Joint Deep Learning for land cover and land use classification

Zhang

Sargent

Pan

et al. 2019

Remote Sensing of Environment

364

163

View full text Add to dashboard Cite

Land cover (LC) and land use (LU) have commonly been classified separately from remotely sensed imagery, without considering the intrinsically hierarchical and nested relationships between them. In this paper, for the first time, a highly novel joint deep learning framework is proposed and demonstrated for LC and LU classification. The proposed Joint Deep Learning (JDL) model incorporates a multilayer perceptron (MLP) and convolutional neural network (CNN), and is implemented via a Markov process involving iterative updating. In the JDL, LU classification conducted by the CNN is made conditional upon the LC probabilities predicted by the MLP. In turn, those LU probabilities together with the original imagery are re-used as inputs to the MLP to strengthen the spatial and spectral feature representations. This process of updating the MLP and CNN forms a joint distribution, where both LC and LU are classified simultaneously through iteration. The proposed JDL method provides a general framework within which the pixel-based MLP and the patch-based CNN provide mutually complementary information to each other, such that both are refined in the classification process through iteration. Given the well-known complexities associated with the classification of very fine spatial resolution (VFSR) imagery, the effectiveness of the proposed JDL was tested on aerial photography of two large urban and suburban areas in Great Britain (Southampton and Manchester). The JDL consistently demonstrated greatly increased accuracies with increasing iteration, not only for the LU classification, but for both 2 the LC and LU classifications, achieving by far the greatest accuracies for each at around 10 iterations. The average overall classification accuracies were 90.18% for LC and 87.92% for LU for the two study sites, far higher than the initial accuracies and consistently outperforming benchmark comparators (three each for LC and LU classification). This research, thus, represents the first attempt to unify the remote sensing classification of LC (state; what is there?) and LU (function; what is going on there?), where previously each had been considered separately only. It, thus, has the potential to transform the way that LC and LU classification is undertaken in future. Moreover, it paves the way to address effectively the complex tasks of classifying LC and LU from VFSR remotely sensed imagery via joint reinforcement, and in an automatic manner.

show abstract

Section: Joint Deep Learning Modelmentioning

confidence: 99%

Joint Deep Learning for land cover and land use classification

Zhang

Sargent

Pan

et al. 2019

Remote Sensing of Environment

364

163

View full text Add to dashboard Cite

show abstract

“…It defines the task of partitioning an image into regions that delineate meaningful objects and labelling those regions with an object label. While it is very popular in computer vision (Ladický et al, 2010, Arbeláez et al, 2012, Chen et al, 2015, it has been barely addressed in the remote sensing community (Montoya-Zegarra et al, 2015, Zheng andWang, 2015). Segmentation segmentation frameworks have demonstrated their usefulness in particular in structured environments such as urban areas.…”

Section: Semantic Segmentation Is a Suitable Solutionmentioning

confidence: 99%

Semantic Segmentation of Forest Stands of Pure Species as a Global Optimization Problem

Dechesne¹,

Mallet²,

Bris³

et al. 2017

ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

ABSTRACT:Forest stand delineation is a fundamental task for forest management purposes, that is still mainly manually performed through visual inspection of geospatial (very) high spatial resolution images. Stand detection has been barely addressed in the literature which has mainly focused, in forested environments, on individual tree extraction and tree species classification. From a methodological point of view, stand detection can be considered as a semantic segmentation problem. It offers two advantages. First, one can retrieve the dominant tree species per segment. Secondly, one can benefit from existing low-level tree species label maps from the literature as a basis for high-level object extraction. Thus, the semantic segmentation issue becomes a regularization issue in a weakly structured environment and can be formulated in an energetical framework. This papers aims at investigating which regularization strategies of the literature are the most adapted to delineate and classify forest stands of pure species. Both airborne lidar point clouds and multispectral very high spatial resolution images are integrated for that purpose. The local methods (such as filtering and probabilistic relaxation) are not adapted for such problem since the increase of the classification accuracy is below 5%. The global methods, based on an energy model, tend to be more efficient with an accuracy gain up to 15%. The segmentation results using such models have an accuracy ranging from 96% to 99%.

show abstract

“…This issue contains three papers concerning segmentation. In [15], an object-based Markov random field model is proposed for semantic segmentation of VHR images. Region size and edge information are employed in the model.…”

Section: Segmentationmentioning

confidence: 99%