A Data Augmentation Strategy Based on Simulated Samples for Ship Detection in RGB Remote Sensing Images

Yan, Yiming; Tan, Zhichao; Su, Nan

doi:10.3390/ijgi8060276

Cited by 36 publications

(18 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To verify the effectiveness of our method, we compared it to Yan's [27] method currently published. Yan et al first proposed the importance of a simulation dataset in remote sensing object detection.…”

Section: Comparison With Other Methodsmentioning

confidence: 99%

Target Detection Based on Simulated Image Domain Migration

Wang

Zhao

et al. 2020

IEEE Access

View full text Add to dashboard Cite

Annotating a large amount of data manually for supervised learning is an indispensable and expensive part. A novel system using the simulation dataset is proposed in this paper. This framework can train the neural networks for remote sensing object detection without any manually labeled dataset. The whole system can be divided into three parts. The first part is the dataset simulator. The simulator synthesizes remote sensing images with the aircraft targets based on real remote sensing images (without any aircraft targets). In the process of data generation, the simulator automatically marks the position information of the aircraft. The second part is the image dataset domain adaptation work. We introduce the work of Cycle-GAN into this part to bridge the perceptual gap between the simulation dataset and reality dataset. Specially, we propose a multi-scale generator into the original Cycle-GAN model to achieve better domain adaptation performance. The final part is the object detection neural network. The domain adaptation quality of the remote sensing images reconstructed by our novel cycle-gan network achieves better performance both in the structural similarity measure and visual appearance. The object detection model trained with the dataset processed by our novel system can get better detection precision. The analytic experiments on the test dataset demonstrate that the object detection model trained with the dataset processed by our novel system can get better detection precision. INDEX TERMS Remote sensing image, generative and adversarial network, object detection, data simulation.

show abstract

Section: Comparison With Other Methodsmentioning

confidence: 99%

Target Detection Based on Simulated Image Domain Migration

Wang

Zhao

et al. 2020

IEEE Access

View full text Add to dashboard Cite

show abstract

“…In the deep learning method, fewer training samples may result in insufficient learning performance of the model, poor generalization of the model and overfitting. To overcome the problem of insufficient sample data in the experiment, we apply the data augmentation [43,44] method of rotation and mirroring to obtain additional similar sample data. As presented in Table 3, the mirroring transformation doubles the number of original sample data.…”

Section: Data Augmentation and Balancementioning

confidence: 99%

A Postearthquake Multiple Scene Recognition Model Based on Classical SSD Method and Transfer Learning

Chen

Yang

et al. 2020

IJGI

View full text Add to dashboard Cite

The recognition of postearthquake scenes plays an important role in postearthquake rescue and reconstruction. To overcome the over-reliance on expert visual interpretation and the poor recognition performance of traditional machine learning in postearthquake scene recognition, this paper proposes a postearthquake multiple scene recognition (PEMSR) model based on the classical deep learning Single Shot MultiBox Detector (SSD) method. In this paper, a labeled postearthquake scenes dataset is constructed by segmenting acquired remote sensing images, which are classified into six categories: landslide, houses, ruins, trees, clogged and ponding. Due to the insufficiency and imbalance of the original dataset, transfer learning and a data augmentation and balancing strategy are utilized in the PEMSR model. To evaluate the PEMSR model, the evaluation metrics of precision, recall and F1 score are used in the experiment. Multiple experimental test results demonstrate that the PEMSR model shows a stronger performance in postearthquake scene recognition. The PEMSR model improves the detection accuracy of each scene compared with SSD by transfer learning and data augmentation strategy. In addition, the average detection time of the PEMSR model only needs 0.4565s, which is far less than the 8.3472s of the traditional Histogram of Oriented Gradient + Support Vector Machine (HOG+SVM) method.

show abstract

“…In the current era of deep learning, RTMs are also strong candidates in the perspectives of data augmentation [50] to alleviate challenges related to data sparsity in order to understand interactions between vegetation and solar radiation, to test potential and limits of methods, and to prepare future missions. DART (Discrete Anisotropic Radiative Transfer) [51,52] is currently one of the most comprehensive 3D RTMs for simulating remotely sensed data, in particular in the context of forest studies [42,53].…”

Section: Introductionmentioning

confidence: 99%

Simulating Imaging Spectroscopy in Tropical Forest with 3D Radiative Transfer Modeling

et al. 2021

View full text Add to dashboard Cite

Optical remote sensing can contribute to biodiversity monitoring and species composition mapping in tropical forests. Inferring ecological information from canopy reflectance is complex and data availability suitable to such a task is limiting, which makes simulation tools particularly important in this context. We explored the capability of the 3D radiative transfer model DART (Discrete Anisotropic Radiative Transfer) to simulate top of canopy reflectance acquired with airborne imaging spectroscopy in a complex tropical forest, and to reproduce spectral dissimilarity within and among species, as well as species discrimination based on spectral information. We focused on two factors contributing to these canopy reflectance properties: the horizontal variability in leaf optical properties (LOP) and the fraction of non-photosynthetic vegetation (NPVf). The variability in LOP was induced by changes in leaf pigment content, and defined for each pixel based on a hybrid approach combining radiative transfer modeling and spectral indices. The influence of LOP variability on simulated reflectance was tested by considering variability at species, individual tree crown and pixel level. We incorporated NPVf into simulations following two approaches, either considering NPVf as a part of wood area density in each voxel or using leaf brown pigments. We validated the different scenarios by comparing simulated scenes with experimental airborne imaging spectroscopy using statistical metrics, spectral dissimilarity (within crowns, within species, and among species dissimilarity) and supervised classification for species discrimination. The simulation of NPVf based on leaf brown pigments resulted in the closest match between measured and simulated canopy reflectance. The definition of LOP at pixel level resulted in conservation of the spectral dissimilarity and expected performances for species discrimination. Therefore, we recommend future research on forest biodiversity using physical modeling of remote-sensing data to account for LOP variability within crowns and species. Our simulation framework could contribute to better understanding of performances of species discrimination and the relationship between spectral variations and taxonomic and functional dimensions of biodiversity. This work contributes to the improved integration of physical modeling tools for applications, focusing on remotely sensed monitoring of biodiversity in complex ecosystems, for current sensors, and for the preparation of future multispectral and hyperspectral satellite missions.

show abstract

A Data Augmentation Strategy Based on Simulated Samples for Ship Detection in RGB Remote Sensing Images

Cited by 36 publications

References 19 publications

Target Detection Based on Simulated Image Domain Migration

Target Detection Based on Simulated Image Domain Migration

A Postearthquake Multiple Scene Recognition Model Based on Classical SSD Method and Transfer Learning

Simulating Imaging Spectroscopy in Tropical Forest with 3D Radiative Transfer Modeling

Contact Info

Product

Resources

About