KR102067242B1 - Method for detecting oil spills on satellite sar images using artificial neural network - Google Patents

Abstract

The present invention relates to an oil leakage detection method using a satellite SAR-based artificial neural network that detects an oil leak per pixel by applying an artificial neural network (ANN) method based on a SAR image, which is a high-resolution all-weather active microwave sensor. Obtaining a SAR image by examining a spatial distribution of an oil distribution using a satellite SAR; And investigating the temporal dispersion of the oil distribution from the acquired SAR image, wherein the investigating the temporal dispersion of the oil distribution comprises: specifying a constant size of a SAR image window, and setting the size of the set window. Designating (NxN) and calculating an average value (m), a standard deviation (std) and an average contrast ratio (σ); Normalizing a designated window of the SAR image; Removing artifacts from the normalized SAR image by using backscatter coefficient attenuation; Determining each pixel in the SAR image as one of oil spill and non-oil according to the calculated information; And outputting an oil outflow distribution result corresponding to the oil outflow range, comprising: identifying pixels corresponding to ships and artifacts and removing incident angle effects successfully from high resolution all-weather wide-area satellite SAR images using the neural network method As a result, oil spills can be detected.

Description

Oil spill detection method using satellite SAR image-based neural network {METHOD FOR DETECTING OIL SPILLS ON SATELLITE SAR IMAGES USING ARTIFICIAL NEURAL NETWORK}

The present invention relates to a method for detecting oil leakage based on satellite SAR (Synthetic Aperture Radar) using artificial neural networks. The present invention relates to a method of detecting an oil spill using a satellite SAR-based neural network that detects an oil spill by pixel.

Oil spills have long occurred as a result of operational failures on ships, accidents on ships, accidents on platforms or other types of accidents around the world's oceans. Numerous major oil spills have occurred, including the 2009 Deepwater Horizon incident, the 1989 Exxon-Valdez spill, and the Hebei Spirit spill from Korea in 2007. . All of these events constituted a major disaster, especially in coastal areas, and their effects varied from short-term deterioration of marine products to long-term damage to the marine environment. Oil spilled by various chemicals and toxic substances can cause huge damage to marine plants and animals, including small fish, coral reefs and mangroves.

Ship and aircraft surveillance is often used to identify the cause of emissions and to monitor oil spills during the cleanup process. In most cases, however, accidents occur in adverse weather conditions that make ship and aircraft difficult to operate. In addition, there are limitations of time delay and narrow space coverage considering the platform operation time for platform movement and wide range observation after accident occurrence.

In contrast, satellite remote sensing, including visible and microwave frequency ranges, can provide an opportunity to overcome these spatial and temporal constraints. As observation opportunities and satellite spatial resolutions increase, oil spill detection using satellite data can play an important role in immediately identifying widespread oil contaminated areas. Unlike human-based observations using ships and aircraft, satellite images are close to real-time maps of large-scale surface runoff oil distributions.

In particular, since SAR is an active microwave sensor, it can provide high quality images of high resolution in an all-weather atmosphere. Considering that marine accidents occur mainly in instrumental conditions, SAR observation images, which are all-weather microwave sensors, are very suitable for rapid and accurate oil spill detection, unlike visible light images that cannot be observed due to weather conditions. Since the oil present in the sea significantly reduces the roughness of the sea surface generated by the wind, it is possible to detect the oil spill region in the image by using the backscatter coefficient attenuation characteristic. Many attempts have been made to develop an algorithm for detecting oil spills using statistical characteristics (mean value (m), standard deviation (σ)) in the window to set the threshold of this attenuation range, and based on this adaptive threshold method (adaptive threshold method) methods, bimodal histogram methods, and the like have been widely used to date. These techniques have the advantage that the operation speed is very fast and no prior information is required, but there is a problem that the detection range varies depending on the set window size and the size or shape of the oil spill area.

Republic of Korea Patent Publication No. 10-1489891 (Invention name: Ocean oil and dangerous hazardous substance spill detection device using indium tin oxide thin film sensor)

Therefore, the present invention is to solve the above-mentioned problems of the prior art, an object of the present invention is to identify the pixels corresponding to the vessel and the artifact and to remove the angle of incidence effect for the successful oil outflow detection by using a high-resolution neural network method To provide an oil spill detection method using a satellite SAR image-based neural network capable of detecting oil spills in a wide-area satellite SAR image.

Another object of the present invention is to recognize and track the temporal movement of oil area and location by comparing satellite image and numerical model data to detect oil more effectively by reflecting the atmospheric and marine environment affecting the evolution of oil spill. It is to provide a method of detecting oil spill using a satellite SAR image that can be produced.

An oil spill detection method using a satellite satellite image-based artificial neural network (SAR), comprising: obtaining a SAR image by examining a spatial distribution of an oil distribution using a satellite SAR; And examining the temporal dispersion of the oil distribution from the acquired SAR image.

Examining the temporal variance of the oil distribution, specifies the size of the SAR image constant window, specifies the size of the set window (NxN), calculates the average value (m), standard deviation (std) and average contrast ratio (σ) Doing; Normalizing a designated window of the SAR image; Removing artifacts from the normalized SAR image by using backscatter coefficient attenuation; Determining each pixel in the SAR image as one of oil spill and non-oil according to the calculated information; And outputting an oil outflow distribution result corresponding to the oil outflow range.

Determining each pixel as one of oil spill and non-oil includes: determining a dark spot for each pixel in the SAR image; And distinguishing noise in the dark spots.

The determining of the dark spot in the SAR image may include: determining a threshold value from a standard deviation (std) of the pixel; A first determination step of determining whether a normalized radar cross section (NRCS) value of the pixel is smaller than a value obtained by subtracting a threshold value k from the average contrast ratio σ; Determining that the NRCS value of the pixel is less than a value obtained by subtracting a threshold value from the average contrast ratio? Determining that the NRCS value of the pixel is equal to or greater than a value obtained by subtracting a threshold value from the average contrast ratio? And a second determination step of determining whether a determination on oil or non-oil for all pixels of the SAR image is completed.

Distinguishing noise at the dark spot may include clustering dark spots of the pixel; Calculating a cluster size of the dark spot of the pixel; A third determination step of determining whether the calculated cluster is larger than a predetermined minimum cluster size; Determining in the third determination step that oil is outflowed when the cluster is larger than the set minimum cluster size; And determining that the cluster is non-oil if the cluster is smaller than or equal to the set minimum cluster size as determined in the third determination step.

The detected dark spots may be configured to connect and cluster in eight neighboring directions.

에서의 후방 산란 표준화 레이더 단면(

)은 입사각의 코사인 제곱과 관련된다는 두 가지 가정에 따라 정규화시키며, 다음의 식으로 표시하도록 하도록 구성될 수 있다. The normalizing the SAR image is such that, since the amount of power radiated back to the satellite sensor follows the cosine law, the radiation variability as a function of the observed area is cosine dependent, and an arbitrary angle

Scattering Standardized Radar Section in

) Is normalized according to two assumptions that relate to the cosine square of the angle of incidence and can be configured to be expressed by the equation

는 θ의 각도에서의 후방 산란 표준화 레이더 단면값이고,

는 후방 산란값(입사각과 무관)임. here,

Is the backscatter normalized radar cross section at the angle θ,

Is the backscatter value (irrespective of the incident angle).

The radar response characteristic from the backscatter normalized radar cross section may be configured to obtain using the following equation.

은 응답특성을 나타내며,

는 θ의 각도에서의 후방 산란 표준화 레이더 단면값이고,

는 정규화된 이미지에 대한 기준각이다. here,

Represents the response characteristics,

Is the backscatter normalized radar cross section at the angle θ,

Is the reference angle for the normalized image.

The removing of the artifact may be configured to identify and remove pixels exhibiting backscattering at least twice as high as the surface of the sea.

Obtaining a SAR image by examining a spatial distribution of an oil distribution using a satellite SAR; And

Investigating the temporal dispersion of the oil distribution from the acquired SAR image;

Examining the temporal dispersion of the oil distribution,

Designating a SAR image constant window size, designating a set window size (NxN), and calculating an average value (m), a standard deviation (std, σ), and an average contrast ratio (σ) of the window size (nxn);

Normalizing the SAR image;

Removing artifacts from the normalized SAR image by using backscatter coefficient attenuation; And

Determining each pixel in the SAR image as one of oil spill and non-oil according to the calculated information;

And outputting an oil outflow distribution result corresponding to the oil outflow range.

Determining each pixel as either oil spill or non-oil

Determining a dark spot for each pixel in the SAR image;

And distinguishing noise in the dark spots.

The determining of the dark spot in the SAR image may include;

Determining a threshold from the standard deviation (std, σ) of the pixel;

Determining whether a normalized radar cross section (NRCS) value of the pixel is smaller than a value obtained by subtracting a threshold value k from the average contrast ratio σ;

Determining an oil spill when the NRCS value of the pixel is less than a value obtained by subtracting a threshold value from the average contrast ratio?

Determining non-oil if the NRCS value of the pixel is greater than or equal to a value obtained by subtracting a threshold value from the average contrast ratio?

And determining whether oil or non-oil for all pixels of the SAR image is completed.

The step of distinguishing noise in dark spots is

Clustering dark spots of the pixel;

Calculating a cluster size of the dark spot of the pixel;

Determining whether the calculated cluster is larger than a predetermined minimum cluster size;

Determining in the determination step that oil is outflowed when the cluster is larger than the set minimum cluster size;

If it is determined in the determination step and the cluster is less than or equal to the set minimum cluster size, the step of determining as a non-oil considering the noise; may be configured to include.

The detected dark spots may be configured to connect and cluster in eight neighboring directions.

Normalizing the SAR image may include:

)을 정규화시키고, Backscatter normalized radar cross-section using squared cosine correction

),

에서의 후방 산란 표준화 레이더 단면(

)은 입사각의 코사인 제곱과 관련된다는 두 가지 가정에 따라 정규화시키며, 다음의 식으로 표시하도록 하도록 구성될 수 있다. The amount of power radiated back to the satellite sensor obeys the cosine law and the radiation variability as a function of the observed area is cosine dependent, and at any angle

Scattering Standardized Radar Section in

) Is normalized according to two assumptions that relate to the cosine square of the angle of incidence and can be configured to be expressed by the equation

는 θ의 각도에서의 후방 산란 표준화 레이더 단면값이고,

는 후방 산란값(입사각과 무관)임. here,

Is the backscatter normalized radar cross section at the angle θ,

Is the backscatter value (irrespective of the incident angle).

It can be configured to obtain the radar response characteristic from the backscatter normalized radar cross section using the following equation.

은 응답특성을 나타내며,

는 의 각도에서의 후방 산란 표준화 레이더 단면값이고,

는 정규화된 이미지에 대한 기준각이다. here,

Represents the response characteristics,

Is the backscatter normalized radar cross section at the angle of,

Is the reference angle for the normalized image.

Removing the artifact,

It may be configured to identify and remove pixels exhibiting backscattering at least twice as high as the surface present in the sea.

Therefore, the oil spill detection method using the satellite SAR image-based neural network of the present invention successfully detects the oil spill from the high resolution all-weather wide-area satellite SAR image using the neural network method by identifying the pixels corresponding to the vessels and the artifacts and removing the incident angle effect. There is a detectable effect.

The oil spill detection method using the satellite SAR image of the present invention reflects the atmospheric and marine environment affecting the evolution of the oil spill by recognizing and tracking the time movement of the oil area and location by comparing the satellite image and the numerical model data. Therefore, the oil can be detected more effectively.

1 is a view showing the contour, waterway and contour of the depth of the vicinity of the Korean Peninsula at the time of the collision point of the oil tanker Hebei Spirit and the oil tanker Wuyi San hit the oil pipeline according to an embodiment of the present invention.
Figure 2 is a view showing a SAR image measured during the Hebei spirit oil spill accident in accordance with an embodiment of the present invention.
3 is a view illustrating a process of detecting an oil spill using a satellite SAR image-based neural network according to an embodiment of the present invention.
4 is a flow chart illustrating a process of investigating the temporal dispersion of oil distribution from the SAR image of FIG. 3 according to an embodiment of the present invention.
FIG. 5 is a flow chart illustrating in more detail the step of determining each pixel of FIG. 4 as either oil spill or non-oil according to an embodiment of the invention. FIG.
FIG. 6 is a flow chart illustrating in more detail the determining the dark spot for each pixel of FIG. 5 in accordance with an embodiment of the present invention. FIG.
7 is a diagram illustrating a distribution value of normalized radar cross-sectional values of an ASAR image according to an embodiment of the present invention.
8 is a flow chart illustrating in more detail the steps of distinguishing noise for each pixel of FIG. 5 in accordance with an embodiment of the present invention.

1 is a view showing the contours, waterways and contours of the depth of the vicinity of the Korean Peninsula when the collision point of the oil tanker Hebei Spirit and the oil tanker Wuyi San hit the oil pipeline according to an embodiment of the present invention.

Referring to FIG. 1, the west coast of Korea is famous for strong tides and seasonal wind changes such as summer and winter monsoons. It has an average value of about 44m and shows various ocean phenomena related to algae and shallow water. Certain coastal areas near Taean, where the Hebei spirit oil spill occurred, are one of the dominant mixes, primarily due to weekly rebound. In this situation, spilled oil droplets can quickly disperse or significantly disperse. For example, in 1993, when a Braer spill occurred along the Shetland coast, a significant amount of oil (about 44%) was reported to have been injected into the water column under the influence of tides. Therefore, it is assumed that oil spills can be evolved by the effects of prevailing wind and tidal mixing in shallow areas of certain areas.

2 is a view showing a SAR image measured during the Hebei spirit oil spill accident in accordance with an embodiment of the present invention.

Referring to FIG. 2, in the present invention, SAR images of the west coast of Korea taken between December 9 and 15, 2007 were used. In the case of the Hebei Spirit spill event, all SAR images except one ENVISAT image were acquired with a single polarization state. Although ENVISAT images were acquired with full polarization, only single-polarized (VV) SAR images were used to avoid problems associated with mismatches in other detection methods. 2 shows a distribution of normalized radar cross section (NRCS) values.

SAR image data used in the present invention was obtained in C-band or L-band (wavelengths of 5.6 cm to 23.6 cm, respectively) with different polarization states in various imaging modes. Depending on the image acquisition mode, the spatial resolution and swath widths of the SAR images varied from a few meters (less than 10m) up to 100m and from 70km to over 400km.

3 is a diagram illustrating a process of detecting an oil leak using a satellite SAR image-based neural network according to an embodiment of the present invention.

Referring to FIG. 3, the oil spill detection method using the satellite SAR image-based artificial neural network according to the present invention may be performed by using a SAR image collected by a computer or a processor on a sea or land.

First, in step S202, a SAR image is acquired by examining a spatial distribution of an oil distribution using a SAR from a satellite. In the present invention, SAR images taken from ENVISAT satellites are used.

In step S204, the temporal dispersion of the oil distribution is examined from the acquired SAR image.

4 is a flowchart illustrating a process of investigating the temporal dispersion of oil distribution from the SAR image of FIG. 3 according to an embodiment of the present invention.

Referring to FIG. 4, in operation S302, a predetermined window size is designated (NxN) in the SAR image, and an average value (m), a standard deviation (std), and an average contrast ratio (σ) corresponding to the size of the specified window size (NxN) Calculate

)을 이용한다. 보다 상세하게 설명하면, 정규화는 위성 센서로 다시 방사되는 전력의 양은 코사인 법칙을 따르기 때문에 관측된 영역의 함수로서의 방사변이(radiation variability)는 코사인 의존적이고, 임의의 각도θ에서의 후방 산란 표준화 레이더 단면(

)은 입사각의 코사인 제곱과 관련된다는 두 가지 가정에 따른다. 즉, 정규화는 다음의 수학식 1로 표시될 수 있다. In step S304, the specified window of the SAR image is normalized. Normalization uses squared cosine correction to backscatter normalized radar cross-sections (

). More specifically, since normalization follows the cosine law, the amount of power radiated back to the satellite sensor is radiation variability as a function of the observed area, which is cosine dependent and backscatter normalized radar cross section at any angle θ. (

) Is based on two assumptions that are related to the cosine square of the angle of incidence. That is, normalization may be represented by Equation 1 below.

는 θ의 각도에서의 후방 산란 표준화 레이더 단면값이고,

는 후방 산란값(입사각과 무관)이다. here,

Is the backscatter normalized radar cross section at the angle θ,

Is the backscatter value (regardless of the incident angle).

In addition, the radar response characteristic from the backscatter normalized radar cross section may be represented by the following equation (2).

은 응답특성을 나타내며,

는 θ의 각도에서의 후방 산란 표준화 레이더 단면값이고,

는 정규화된 이미지에 대한 기준각이다. here,

Represents the response characteristics,

Is the backscatter normalized radar cross section at the angle θ,

Is the reference angle for the normalized image.

In step S306, the artifact is removed by using the backscattering attenuation characteristic corresponding to the designated window of the normalized SAR image. Removal of artifacts identifies and removes pixels exhibiting backscattering that is greater than or equal to the surface of the sea surface. For example, pixels that generate backscattering more than twice as high as the sea level may be regarded as artifacts and removed.

In operation S308, the controller determines one of the oil spill and the non-oil for each pixel of the corresponding window of the SAR image according to the calculated information.

In step S310 outputs an oil outflow distribution result corresponding to the oil outflow range. Since the oil spill distribution result is output in pixels, it can show very high accuracy.

FIG. 5 is a flow chart illustrating in more detail a step of determining each pixel of FIG. 4 as one of oil spill and non-oil according to an embodiment of the present invention.

Referring to FIG. 5, in operation S402, a dark spot for each pixel is determined within a corresponding window of the SAR image.

In step S404, noise is distinguished from the dark spot. That is, when the dark spot is caused by the noise, the dark spot is determined by the noise, and accordingly, pixels that are incorrectly determined to be oil spills are classified as non-oil, thereby eliminating the influence of the noise.

FIG. 6 is a flow chart illustrating in detail the step of determining the dark spot for each pixel of FIG. 5 in accordance with an embodiment of the present invention.

Referring to FIG. 6, in operation S502, the threshold k is determined from the standard deviation std of the pixel. The threshold value k is determined based on the ensemble mean normalized radar cross section (NRCS).

In step S504, it is determined whether a normalized radar cross section (NRCS) value of the pixel is smaller than a value obtained by subtracting a threshold value k from the average contrast ratio σ.

7 is a diagram illustrating a neural network structure for determining a dark spot for each pixel of FIG. 5 using a neural network according to another exemplary embodiment of the present invention.

Referring to FIG. 7, in the present invention, a two-layer feedforward backpropagation neural network is used to detect outflow by pixels to determine dark spots.

The input layer consists of two values of NRCS and two sets of wind speeds and six texture values estimated from the CMOD5 algorithm at a given SAR image pixel, namely, the average intensities N_1 × N_1 (Window 1) and N_2 × N_2 (Windows 2). At different window sizes, it represents the average brightness (m), average contrast ratio (σ), smoothness (R), third moment (μ), uniformity (U), and entropy (e) at the center pixel, with each texture parameter: It is determined by the equation (3).

Where zi is the pixel value, p (zi) is the histogram of the NRCS in the window, and L is the number of NRCS levels. Boundary window sizes were determined to be 21 × 21 and 51 × 51.

Network optimization was performed using the Levenberg-Marquardt method. In each learning process, the weights and biases were updated and minimized using the mean squared error. Many neurons in each layer can be determined based on trial and error and computational efficiency. According to the experimental results, the number of neurons in the hidden and output layers was set to 5 and 1, respectively.

To build the training function, four SAR images including dark dark spots and backgrounds (as shown in FIGS. 2B, 2D, 2E and 2F) were selected and databased into training sets. The training process was performed using the training data to update the weights and deviations of the learning function until the mean squared error reached the performance target of 0.1.

8 is a diagram illustrating a distribution value of normalized radar cross-sectional values of an ASAR image according to an embodiment of the present invention.

Referring to FIG. 8, (a) of FIG. 8 illustrates an NRCS distribution of an Envisat ASAR image acquired at 01:40, Universal Time Coordinated (UTC) on December 11, 2007. 8B shows the detection result by the adaptive threshold method. 8C shows the detection result by the neural network method. 8 (d) shows the difference between the result of the adaptive threshold method and the neural network method.

In step S506, when the NRCS value of the pixel is smaller than the value obtained by subtracting the threshold value k from the average contrast ratio sigma (step S504, Yes), it is determined as an oil spill. That is, as described above, an NRCS pixel having a low average contrast ratio compared to a threshold value is determined as a 'dark spot'.

If the NRCS value of the pixel is greater than or equal to the value obtained by subtracting the threshold value k from the average contrast ratio sigma (S508, No), it is determined to be non-oil.

In operation S510, it is determined whether an oil or non-oil determination for all pixels of the SAR image is completed. If the determination of the oil or non-oil for all the pixels of the SAR image is completed (step S510, Yes) and returns to step S512 of FIG. 9 to be described later, the determination result is the oil or non-oil for all pixels of the SAR image If the determination is not completed (step S510, No), the process returns to step S502.

9 is a flow chart illustrating in more detail the steps of distinguishing noise for each pixel of FIG. 5 in accordance with an embodiment of the present invention.

9, in operation S512, dark spots of the pixels are clustered. Clustering may be configured to connect and cluster the detected dark spots in eight neighboring directions.

In operation S514, the cluster size of the dark spot of the pixel is calculated.

In operation S516, it is determined whether the calculated size S _clust is greater than a predetermined size S _{min of the} cluster.

If the calculated cluster size (S _clust ) is larger than the set minimum cluster size (S _min ) (step S516, Yes), it is determined as oil leakage (step S518). That is, it is determined that the cluster is not affected by noise.

If the calculated cluster size S _clust is less than or equal to the set minimum cluster size S _min (step S516, No), the noise is determined as non-oil.

On the other hand, in the detailed description, the dark spot and the oil spill are used in the same sense, and the dark spot represents a state that appears as a result of analyzing the SAR image. That is, when interpreting the SAR image, the corresponding pixel or cluster may be determined as a dark spot in the image determination step. When the noise is removed according to the present invention, it is determined as an oil spill. Therefore, oil spill and dark spot are used in the same sense, but there are only slight differences in the interpretation criteria.

On the other hand, the method of the present invention can be performed using a computer that acquires the data obtained from the satellite and the sensor installed in the satellite through the computer. Alternatively, the result obtained from the satellite and the sensor installed in the satellite may be input to a processor performing a certain operation and performed by the processor.

Although the contents of the present invention have been described with reference to the embodiments illustrated in the drawings, this is merely exemplary, and it should be understood by those skilled in the art that various modifications and equivalent other embodiments are possible. will be. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

Claims (8)

In the oil spill detection method using a satellite SAR (Synthetic Aperture Radar) image-based artificial neural network,
Obtaining a SAR image by examining a spatial distribution of an oil distribution using a satellite SAR; And
Investigating the temporal dispersion of the oil distribution from the acquired SAR image;
Examining the temporal dispersion of the oil distribution,
Designating a SAR image constant window size, designating a size of the set window (NxN), and calculating an average value (m), a standard deviation (std), and an average contrast ratio (σ);
Normalizing a designated window of the SAR image;
Removing artifacts from the normalized SAR image by using backscatter coefficient attenuation;
Determining each pixel in the SAR image as one of oil spill and non-oil according to the calculated information; And
And outputting an oil spill distribution result corresponding to the oil spill range.
Determining each of the pixels as one of oil spill and non-oil,
Determining a dark spot for each pixel in the SAR image;
And discriminating noise in the dark spots;
Determining a dark spot from the SAR image,
Determining a threshold from the standard deviation (std) of the pixel;
A first determining step of determining whether a normalized radar cross section (NRCS) value of the pixel is smaller than a value obtained by subtracting a threshold value k from the average contrast ratio?
Determining that the NRCS value of the pixel is less than a value obtained by subtracting a threshold value from the average contrast ratio?
Determining that the NRCS value of the pixel is equal to or greater than a value obtained by subtracting a threshold value from the average contrast ratio? And
And a second determination step of determining whether the determination of the oil or non-oil for all the pixels of the SAR image is completed. delete delete The method of claim 1, wherein the discriminating noise in the dark spots comprises:
Clustering dark spots of the pixel;
Calculating a cluster size of the dark spot of the pixel;
A third determination step of determining whether the calculated cluster is larger than a predetermined minimum cluster size;
Determining in the third determination step that oil is outflowed when the cluster is larger than the set minimum cluster size; And
And determining that the cluster is non-oil if the cluster is smaller than or equal to the set minimum cluster size as determined in the third determination step, and determining non-oil as the noise. The method of claim 4, wherein the detected dark spots,
An oil spill detection method using a satellite SAR image-based neural network for clustering by connecting to eight neighboring directions. The method of claim 1, wherein normalizing the SAR image comprises:
Since the amount of power radiated back to the satellite sensor follows the cosine law, the radiation variability as a function of the observed area is cosine dependent,

Scattering Standardized Radar Section in

) Is normalized according to two assumptions that are related to the cosine square of the angle of incidence, and is configured to be expressed by the following equation.

here,

Is the backscatter normalized radar cross section at the angle θ,

Is the backscatter value (irrespective of the incident angle). The method of claim 6, wherein the radar response characteristic from the backscatter normalized radar cross section,
Oil leakage detection method using a satellite SAR image-based artificial neural network that is configured to obtain using the following equation.

here,

Represents the response characteristics,

Is the backscatter normalized radar cross section at the angle of,

Is the reference angle for normalized images The method of claim 1, wherein the removing the artifact,
An oil spill detection method using a satellite SAR image-based neural network, which is configured to identify and remove pixels exhibiting backscattering at least twice as high as the surface existing in the sea.

Images