Deep learning-based hyperspectral oil spill detection for marine pollution monitoring in the Gulf of Mexico: A step toward marine pollution monitoring and SDG 14 compliance.

Oil spills cause a significant threat to marine and coastal ecosystems. It is one of the major causes of water pollution. This research focuses on the use of deep learning for oil spills detection and classification. UNet is a convolutional neural network, originally proposed for biomedical image segmentation and modified for the discrimination of oil spills and look-alikes. The model is trained on a publicly available benchmark oil spill detection dataset of Sentinel-1 synthetic aperture radar (SAR) images. The images have been semantically segmented into multiple regions of interest such as sea surface, oil spills, look-alikes, ships and land. The proposed UNet-based model achieves intersection over union (IoU) value of 95.69% for sea surface, 60.85% for oil spills, 54.90% for look-alikes, 70.27% for ships and 96.79% for land class. The mean intersection over union (mIoU) value for all the classes is 75.70% which consitutes a nearly 10% increase compared to state of the art for this dataset.

Marine oil spill is a major threat to marine and coastal ecosystems and is seen relatively often, such as the Deepwater Horizon oil spill disaster in the Gulf of Mexico in 2010 and Bohai Sea oil spills in China in 2011. Fast and accurate discrimination of oil spill is the largest challenge in detection of oil spills using remote sensing technology. In this research imaging spectroscopic analysis and Oil Slope Index(OSI) were developed to map the locations of surface crude oil in Gulf of Mexico using the SpecTIR data which was collected at 2.2m GSD and 360 spectral channels, covering 390–2450nm. The spectral features and differences of the main objects of oil, sea water and clouds can be found in the DN value of pixel spectra. The slope difference in the range from 550nm to 750nm between crude oil and other objects can be taken as a key feature for detection of crude oil on the sea surface. The Oil Slope Index(OSI) avoids the absorption bands of O <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> and H <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> O in the air and transforms the imaging spectroscopy data into a single band image that shows the distribution of crude oil spill. OSI values can be easily calculated from radiance or DN data and no additional pre-processing of the imagery was necessary before crude oil detection. The result shows that the algorithms work well for oil spill detection which integrated the spectral feature of oil, sea water and clouds by establishing a decision tree. The automatic determination of thresholds by applying Otsu's image segmentation can realize the fast and automatic extraction of surface crude oil. This study demonstrated that the Oil Slope Index (OSI) has the potential to become a useful image processing algorithm and operational tool for imaging spectroscopy detection of crude oil spill.

A self-evolving deep learning algorithm for automatic oil spill detection in Sentinel-1 SAR images

Oil spillage over a sea or ocean surface is a threat to marine and coastal ecosystems. Spaceborne synthetic aperture radar (SAR) data have been used efficiently for the detection of oil spills due to their operational capability in all-day all-weather conditions. The problem is often modeled as a semantic segmentation task. The images need to be segmented into multiple regions of interest such as sea surface, oil spill, lookalikes, ships, and land. Training of a classifier for this task is particularly challenging since there is an inherent class imbalance. In this work, we train a convolutional neural network (CNN) with multiple feature extractors for pixel-wise classification and introduce a new loss function, namely, “gradient profile” (GP) loss, which is in fact the constituent of the more generic spatial profile loss proposed for image translation problems. For the purpose of training, testing, and performance evaluation, we use a publicly available dataset with selected oil spill events verified by the European Maritime Safety Agency (EMSA). The results obtained show that the proposed CNN trained with a combination of GP, Jaccard, and focal loss functions can detect oil spills with an intersection over union (IoU) value of 63.95%. The IoU value for sea surface, lookalikes, ships, and land class is 96.00%, 60.87%, 74.61%, and 96.80%, respectively. The mean intersection over union (mIoU) value for all the classes is 78.45%, which accounts for a 13% improvement over the state of the art for this dataset. Moreover, we provide extensive ablation on different convolutional neural networks (CNNs) and vision transformers (ViTs)-based hybrid models to demonstrate the effectiveness of adding GP loss as an additional loss function for training. Results show that GP loss significantly improves the mIoU and F1 scores for CNNs as well as ViTs-based hybrid models. GP loss turns out to be a promising loss function in the context of deep learning with SAR images.

Oil spills are one of the most dangerous catastrophes that threaten the oceans. Therefore, detecting and monitoring oil spills by means of remote sensing techniques that provide large-scale assessments is of critical importance to predict, prevent, and clean oil contamination. In this study, the detection of an oil spill using synthetic aperture radar (SAR) imagery is considered. Detection of the oil spill is performed using change detection algorithms between imagery acquired at different times. The specific algorithms used are the correlation coefficient change statistic and the intensity ratio change statistic algorithms. These algorithms and the probabilistic selection of threshold criteria are reviewed and discussed. A recently offered change detection method that depends on generating change maps of two images in a temporal sequence is used. An initial change map is obtained by cumulatively adding sequences in such a manner that common change areas are excluded and uncommon change areas are included. A final change map is obtained by comparing the first and the last images in the temporal sequence. This method requires at least three images to be employed and can be generalized to longer temporal image sequences. The purpose of this approach is to provide a double-check mechanism to the conventional approach and, thus, reduce the probability of false alarm while enhancing change detection. The algorithms are tested on 2010 Gulf of Mexico oil spill imagery. It is shown that the intensity ratio change statistic is a better tool for identification of the changes due to the oil spill compared to the correlation coefficient change statistic. It is also shown that the proposed method can reduce the probability of false alarm.

Detected oil spills are usually classified according to confidence levels. Such levels are supposed to describe the probability that an observed dark feature in the satellite image is related to the actual presence of an oil spill. The Synthetic Aperture Radar (SAR) derived oil spill detection probability estimation has been explored as an intrinsic aspect of oil spill classification, which fundamentally computes the likelihood that the detected dark area is related to an oil spill. However, the SAR based probability estimation should be integrated with additional criteria in order to become a more effective tool for the End Users. As example, the key information for the final users is not the confidence level of the detection “per se” but the alert (i.e. the potential impact of the pollution and the possibility to catch the polluter red-handed) that such detection generates. This topic was deeply discussed in the framework of the R and D European Group of Experts on remote sensing Monitoring of marine Pollution (EGEMP) and a paper was published in 2010. The newly established EMSA CleanSeaNet service (2nd generation) provides the alert level connected to the detection of a potential oil spill in a satellite image based on the likelihood of being an oil spill in combination with impact and culprit information.

A novel deep learning method for marine oil spill detection from satellite synthetic aperture radar imagery

Oil spill detection techniques using various decomposition algorithms have been studied by researchers and are still evolving. In this paper, two oil spill detection approaches based on Polarimetric decomposition are studied and a combination of useful features is used for efficient oil spill detection and differentiation. The first approach is based on coherent decomposition algorithm i.e. Pauli Decomposition and the second is based on the incoherent decomposition algorithm i.e. H/A/\(\alpha \) decomposition. The H/A/\(\alpha \) decomposition detects the different types of weathered oil spill but fails to discriminate the oil spill from look alike. Pauli decomposition detects and discriminates oil spill from the look-alike but fails to discriminate the type of oil spill. An improved technique is proposed to detect and characterize the type of weathered oil using the combination of the features of H/A/\(\alpha \) and Pauli decomposition. The proposed approach is implemented and validated using the L band UAVSAR data acquired from the Deepwater Horizon Oil spill at Gulf of Mexico in June 2010. The accuracy analysis of the proposed approach using the SVM classification shows that the proposed combination not only detects the oil spill patches but also distinguishes the type of weathered oil with higher accuracy as compared to individual approaches.

A typical oil spill recovery vessel has been historically outfitted with an oil spill detection (OSD) radar. During an oil spill recovery operation, there is a dedicated operator who is responsible for interpreting information from the radar image. Industry developments over the last several years now require that an OSD radar automatically detect and track an oil spill. There are two primary needs driving this development. The first is that OSD systems and operations are becoming more sophisticated; automatic OSD aids for a more efficient oil spill operation where an operator's attention may be directed to a potential spill. The automatic OSD also aids a multi-sensor system; one such example is where an OSD radar is used to steer an IR camera to a candidate spill for more detailed evaluation or validation. The other primary driver for automatic OSD is for monitoring systems, which serve for early warning. Monitoring systems may be found along coastal installations or oil platforms. The automatic spill detection functionality of an OSD system may be implemented in different levels of sophistication. Perhaps the simplest configuration is one that uses fixed thresholds relative to the image for alarming whether a region in a radar image is a spill or not. The benefit of simple threshold detector is that it is easy to implement in software. The weakness is that it is prone to both lower overall detection rate and high false alarm rate. A more robust automatic spill detection method is one that treats it as an image-processing problem. The paper here presents a model based OSD. Generation of confidence maps is central to the method and provides an indication of the likelihood of oil. Inputs to the confidence maps come from multiple sources, several of which are based on uniquely constructed models. Among these is a histogram comparator, which scans a radar image and compares the data to reference models from real oil spills. A discussion of the methods used focuses on (a) the necessary steps prior to the confidence map construction, (b) how the confidence maps are layered with inputs, (c) how the information in the confidence maps is transitioned into the detection of oil, (d) and finally alarming.

Monitoring offshore oil pollution using multi-class convolutional neural networks

Oil spill or leakage into waterways and ocean spreads very rapidly due to the action of wind and currents. The study of the behavior and movement of these oil spills in sea had become imperative in describing a suitable management plan for mitigating the adverse impacts arising from such accidents. But the inherent difficulty of discriminating between oil spills and look-alikes is a main challenge with Synthetic Aperture Radar (SAR) satellite data and this is a drawback, which makes it difficult to develop a fully automated algorithm for detection of oil spill. As such, an automatic algorithm with a reliable confidence estimator of oil spill would be highly desirable. The main objective of this work is to develop comparative automatic detection procedures for oil spill pixels in multimode (Standard beam S2, Wide beam W1 and fine beam F1) RADARSAT-1 SAR satellite data that were acquired in the Malacca Straits using two algorithms namely, post supervised classification, and neural network (NN) for oil spill detection. The results show that NN is the best indicator for oil spill detection as it can discriminate oil spill from its surrounding such as look-alikes, sea surface and land. The receiver operator characteristic (ROC) is used to determine the accuracy of oil spill detection from RADARSAT-1 SAR data. The results show that oil spills, lookalikes, and sea surface roughness are perfectly discriminated with an area difference of 20% for oil spill, 35% look–alikes, 15% land and 30% for the sea roughness. The NN shows higher performance in automatic detection of oil spill in RADARSAT-1 SAR data as compared to Mahalanobis classification with standard deviation of 0.12. It can therefore be concluded that NN algorithm is an appropriate algorithm for oil spill automatic detection and W1 beam mode is appropriate for oil spill and look-alikes discrimination and detection. Key words: Oil Spill, RADARSAT-1 SAR data, Mahalanobis classification, neural network (NN).

Oil spills is a major threat to ocean ecosystems. The capability of synthetic aperture radar (SAR) sensors to detect oil spills over the sea surface is established and proven. Oil spills appear as dark spots in SAR images. However, dark formations can be caused by a number of phenomena (e.g., manmade actions, geological conditions, and meteorological and hydrological effects). The current researches aims to distinguish oil spills or look-alikes. The methods include three main parts, dark formation detection, feature extraction, and classification. To further improve the accuracy and efficiency of SAR image segmentation in the detection of marine oil spill, this study presents a new framework for oil spill and look-alike classification based on kernel fuzzy C-means (KFCM) and ontology. Ontologies are employed to model semantic concepts about shape, existence time, forming reason, and texture. Pre-classification can be performed by matching each region with the concepts of ontology. Once the needed dark formations have been selected, KFCM algorithm is used for classification. Experiment results show that this method performs well.

Application of C-band sentinel-1A SAR data as proxies for detecting oil spills of Chennai, East Coast of India

Monitoring and M A Record-Breakin Geophysical Mon Copyright 2011 b 10.1029/2011GM Oil pollution is a threat that increasingly concerns marine/coastal ecosystem. Timely detection and continuous update of information are fundamental to reduce oil spill environmental impact. EOSs, especially meteorological satellites, can be profitably used for a near real time sea monitoring thanks to their high temporal resolution and easy data delivery. In this paper, we present a new algorithm, based on the general Robust Satellite Technique (RST) approach, for automatic near-real-time oil spill detection and continuous monitoring (i.e., in both daytime and nighttime) by using optical data. The new RST scheme has been applied to the analysis of the recent oil spill disaster of the Deepwater Horizon Platform in the Gulf of Mexico. In particular, a dense temporal series of RST-based oil spill maps, obtained by using Moderate Resolution Imaging Spectroradiometer-thermal infrared records acquired in both daytime and nighttime during the 25–29 April 2010 period, are shown and commented. The results seem to confirm the good performance of the proposed approach in automatic detection of oil spill presence with a high level of reliability and sensitivity even in nighttime acquisitions. These achievements confirm the potential of optical data for oil spill detection and monitoring, thus suggesting their use in combination with radar acquisitions toward developing a multiplatform system that is able to furnish detailed and frequent information about oil spill presence and dynamics.

Although advancements in remote sensing technology have facilitated quick capture and identification of the source and location of oil spills in water bodies, the presence of other biogenic elements (lookalikes) with similar visual attributes hinder rapid detection and prompt decision making for emergency response. To date, different methods have been applied to distinguish oil spills from lookalikes with limited success. In addition, accurately modeling the trajectory of oil spills remains a challenge. Thus, we aim to provide further insights on the multi-faceted problem by undertaking a holistic review of past and current approaches to marine oil spill disaster reduction as well as explore the potentials of emerging digital trends in minimizing oil spill hazards. The scope of previous reviews is extended by covering the inter-related dimensions of detection, discrimination, and trajectory prediction of oil spills for vulnerability assessment. Findings show that both optical and microwave airborne and satellite remote sensors are used for oil spill monitoring with microwave sensors being more widely used due to their ability to operate under any weather condition. However, the accuracy of both sensors is affected by the presence of biogenic elements, leading to false positive depiction of oil spills. Statistical image segmentation has been widely used to discriminate lookalikes from oil spills with varying levels of accuracy but the emergence of digitalization technologies in the fourth industrial revolution (IR 4.0) is enabling the use of Machine learning (ML) and deep learning (DL) models, which are more promising than the statistical methods. The Support Vector Machine (SVM) and Artificial Neural Network (ANN) are the most used machine learning algorithms for oil spill detection, although the restriction of ML models to feed forward image classification without support for the end-to-end trainable framework limits its accuracy. On the other hand, deep learning models’ strong feature extraction and autonomous learning capability enhance their detection accuracy. Also, mathematical models based on lagrangian method have improved oil spill trajectory prediction with higher real time accuracy than the conventional worst case, average and survey-based approaches. However, these newer models are unable to quantify oil droplets and uncertainty in vulnerability prediction. Considering that there is yet no single best remote sensing technique for unambiguous detection and discrimination of oil spills and lookalikes, it is imperative to advance research in the field in order to improve existing technology and develop specialized sensors for accurate oil spill detection and enhanced classification, leveraging emerging geospatial computer vision initiatives.