Convolutional neural network model for discrimination of harmful algal bloom (HAB) from non-HABs using Sentinel-3 OLCI imagery

Shin, Jisun; Khim, Boo‐Keun; Jang, Lee-Hyun; Lim, Jin-Wook; Jo, Young‐Heon

doi:10.1016/j.isprsjprs.2022.07.012

Cited by 13 publications

(4 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Analysing the same dataset using different approaches is a reasonable way to confirm how well a new technique may perform in comparison to previous ones, but this repetition also indicates that similar investigations of various issues is still challenging, probably due to the small data problem. Nevertheless, we found that a few studies have investigated the monitoring of extreme events, including natural fire occurrences (Kato et al, 2021;Zhenyang Xue et al, 2022) and algal bloom events (Shin et al, 2022). We expect future studies to address other equally important global change events, such as conflicts, energy issues, and biodiversity problems.…”

Section: Variousmentioning

confidence: 96%

Ten deep learning techniques to address small data problems with remote sensing

Safonova¹,

Ghazaryan²,

Stiller³

et al. 2023

Preprint

View full text Add to dashboard Cite

Researchers and engineers have increasingly used Deep Learning (DL) for a variety of Remote Sensing (RS) tasks. However, data from local observations or via ground truth is often quite limited for training DL models, especially when these models represent key socio-environmental problems, such as the monitoring of extreme, destructive climate events, biodiversity, and sudden changes in ecosystem states. Such cases, also known as small data problems, pose significant methodological challenges. This review summarises these challenges in the RS domain and the possibility of using emerging DL techniques to overcome them. We show that the small data problem is a common challenge across disciplines and scales that results in poor model generalisability and transferability, yet this has not been investigated in a structured way. We first introduce ten emerging DL techniques: transfer learning, self-supervised learning, semi-supervised learning, few-shot learning, zero-shot learning, active learning, weakly supervised learning, multitask learning, process-aware learning, and ensemble learning; we also include a validation technique known as spatial k-fold cross validation. These techniques have shown promising potential in other scientific disciplines, but have been rarely applied in the RS domain. We also provide guidance on which learning technique to use in various cases, which helps to create a more methodologically robust DL application (and a greater number of them) that can be used to tackle socially important problems with limited data.

show abstract

Section: Variousmentioning

confidence: 96%

Ten deep learning techniques to address small data problems with remote sensing

Safonova¹,

Ghazaryan²,

Stiller³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…However, because GOCI does not have a 510 nm band, we used the OC3 algorithm with common bands at 443, 490, and 555 nm. The GOCI TSM product was estimated using the Yellow Sea and East China Sea Ocean Color (YOC) algorithms [49] in (9).…”

Section: Performance Evaluation Of the U-net Sr Modelmentioning

confidence: 99%

“…CEAN color remote sensing data have been widely used for short-and long-term monitoring of water quality [1], [2], [3], [4], ocean productivity [5], ocean fronts [6], and blooms such as red [7], [8], [9], green, and golden tide blooms [10], [11], [12]. Monitoring and assessing water changes are carried out using ocean color sensors such as the Moderate Resolution Imaging Spectroradiometer (MODIS; Terra:1999-present; Aqua:2002-present), Sea-viewing Wide Manuscript received XXX; revised XXX; accepted XXX.…”

Section: Introductionmentioning

confidence: 99%

“…Neckel et al [24] reported that OLCI is efficient for the geospatial analysis of suspended sediments in terrestrial marine estuaries. Shin et al [9] suggested that OLCI is effective for monitoring red tide blooms off the Korean coast because the Korean Peninsula has complex coastlines and numerous islands. Geostationary KOMPSAT 2B (GK2B) GOCI-II, launched on February 18, 2020, is the successor to the GOCI mission.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

U-Net Super-Resolution Model of GOCI to GOCI-II Image Conversion

Shin,

Jo,

Khim

et al. 2024

IEEE Trans. Geosci. Remote Sensing

Self Cite

View full text Add to dashboard Cite

The use of ocean color sensors presents limitations to monitoring coastal environmental changes and capturing fine spatial features below 1 km due to low spatial (0.5-1 km) and temporal (1 day) resolutions. The Geostationary Ocean Color Imager (GOCI)-II, launched on 18 February 2020, is a follow-up mission to GOCI, operated from June 27, 2010 to March 31, 2021. GOCI-II imagery, with a spatial resolution of 250 m, detects more detailed spatial structures of ocean dynamics compared to GOCI with a spatial resolution of 500 m. This study aims to develop a U-Net super-resolution model to enhance the GOCI remote-sensing reflectance (Rrs) imagery to the same spatial resolution as GOCI-II. The U-Net model is trained with eight paired bands (412, 443, 490, 555, 660, 680, 745, and 865 nm) of GOCI and GOCI-II Rrs around the waters of the Korean Peninsula. The consistency level between GOCI and GOCI-II images indicated GOCI sensor degradation, especially in the blue bands, during its last mission period from December 2020 to March 2021. Through quantitative and qualitative evaluations, we found that the U-Net Rrs image had greater spectral information with higher consistency compared to the G1-bicubic image by bicubic interpolation of GOCI. In particular, the U-Net results improved the consistency in the blue bands (412, 443, and 490 nm). Qualitative evaluations also showed that U-Net corrected the blue band underestimation in degraded GOCI images. In addition, chlorophyll-a concentration (CHL) map from the U-Net Rrs not only simulated spatial patterns, similar to GOCI-II CHL map, but also corrected the overestimated GOCI CHL map. The U-Net super-resolution model may help to produce more reliable and fine-scale Rrs products from GOCI similar to those of GOCI-II, and to enable long-term ocean color monitoring around the waters of the Korean Peninsula.

show abstract