An unmanned aerial vehicle hyperspectral non-imaging paradigm oil slick detection method

Abstract

The application discloses a kind of unmanned aerial vehicle hyperspectral non-imaging paradigm's oil slick detection method, comprising: parallel push-scan type scanning is carried out to offshore oil spill area, and the hyperspectral original data of measurement sea area is obtained;Spectral mixing matrix pixel splicing method is used to seamlessly splice the hyperspectral original data into image to obtain clean sea area data, other coastline and land non-oil slick area hyperspectral pixel cube;Full-spectral data set obtained in oil slick area and suspected oil slick area is regarded as foreground, and corresponding area relative position information is retained, other clean sea area data, other coastline and land non-oil slick area data are marked as background;The background is carried out hyperspectral cube sparse representation, and spectral segment sparse line compression is carried out using matrix mapping mode to obtain sparse data set, and the feature data set of oil slick area and sparse data set are used as transmission data;Containing relative position information oil slick area is marked to background image, and oil slick area focusing result is marked on map.

Description

Technical Field

[0001] This invention relates to the field of oil spill detection technology, and in particular to an oil spill detection method using a hyperspectral non-imaging paradigm employed by unmanned aerial vehicles (UAVs). Background Technology

[0002] The risk of marine oil spills is constantly increasing. Maritime accidents involving oil tankers, such as collisions, groundings, contact, storms, fires, explosions, and sinkings, occur frequently during maritime transport. These oil-related accidents not only cause enormous human and economic losses but also affect the maritime transport of other vessels and even cause serious and lasting damage to the marine environment, drawing widespread attention worldwide. Optical remote sensing technology, by detecting the reflected radiation spectral information of oil-contaminated and clean seawater surfaces, has become an effective means of marine oil spill monitoring. On the one hand, hyperspectral oil spill detection technology can obtain potential characteristic values ​​of reflected radiation intensity at different wavelengths in the oil spill detection area, effectively improving the accuracy of oil spill area detection. On the other hand, data transmission technology in UAV hyperspectral remote sensing is crucial and plays an important role in oil spill detection. Currently, UAV remote sensing technology is developing towards high resolution and multi-sensor platform coexistence, with its temporal resolution, spatial resolution, and spectral resolution all generally improving, leading to a continuous increase in the demand for data transmission. These developments and changes primarily impact data transmission demands in the following ways: UAV hyperspectral remote sensing data contains a large amount of information, requiring high timeliness in data acquisition; secondly, there is a growing demand for improved data security. UAV hyperspectral non-imaging oil spill detection technology can directly process raw reflectance radiation data, bypassing the imaging process to quickly acquire oil spill characteristic areas and directly detect these areas. This enables rapid mapping of oil spill areas, timely and effective location of spill zones, and facilitates rapid implementation of oil spill emergency response strategies by relevant maritime authorities. Summary of the Invention

[0003] To address the problems existing in the prior art, this invention discloses a method for detecting oil slicks using a drone's hyperspectral non-imaging paradigm, specifically comprising the following steps:

[0004] Parallel push-broom scanning was used to acquire raw hyperspectral data of the measured sea area in the oil spill zone.

[0005] A spectral mixing matrix pixel stitching method is used to seamlessly stitch together hyperspectral raw data to obtain hyperspectral pixel cubes of clean sea areas, coastlines, and non-oil-slick areas on land.

[0006] Using hyperspectral pixel cubes as input data, oil spill areas are detected based on the full spectrum features of oil spill, with one or more pixels as units. The full spectrum dataset of oil spill areas and suspected oil spill areas is used as the foreground, and the relative position information of the corresponding areas is preserved. Clean sea area data, coastline and land non-oil spill area data are marked as background.

[0007] The background is represented by a hyperspectral cube sparse representation, and the sparse dataset is obtained by spectral band sparse row compression using matrix mapping. The feature dataset and sparse dataset of the oil spill area are used as the transmission data.

[0008] The oil slick area containing relative location information is marked on the background image, and the focusing result of the oil slick area is plotted on the map.

[0009] Based on existing hyperspectral data of oil spill areas, the full-spectral characteristics of oil spill areas and other locations at the same time are determined. A spectral comparison algorithm is used to compare the spectral bands of the area to be tested, and the detection weight parameters in the spectral comparison algorithm are set to calculate the confidence levels of oil spill areas and suspected oil spill areas in different spectral comparison algorithms.

[0010] The clean sea surface type of the oil spill area and suspected oil spill areas is obtained.

[0011] When obtaining a sparse dataset by performing sparse row compression of spectral bands using matrix mapping: a sparse measurement matrix is ​​constructed and randomly mapped, where the mapping condition satisfies the constraint of isometry. Specifically, the mapping satisfies:

[0012] y = R x Where x represents the high-dimensional data of each measured pixel, and its x-dimensionality is x∈R d Let y represent low-dimensional data obtained by reconstruction using a random mapping matrix, y∈R k The dimension of the corresponding random measurement mapping matrix R is k. d (k≤d), all elements r ij The probability density function is a ~N(0,1) distribution that satisfies:

[0013]

[0014] A spectral band selection affinity evaluation criterion is introduced, which uses an affinity loss function to determine whether the sparse representation vector obtained by spectral band degradation is suitable.

[0015]

[0016] Where X is a low-dimensional matrix obtained through random mapping, D represents the low-dimensional combined feature spectral band, and Z represents the sparse coefficient matrix, which is the iteration coefficient required for spectral matrix compression, thereby achieving the effect of foreground spectral band degradation.

[0017] By adopting the above technical solution, the present invention provides a method for detecting floating oil using a UAV hyperspectral non-imaging paradigm. This method can largely solve the technical bottleneck of receiving hyperspectral remote sensing data information by remote sensing analysis platforms, transforming the traditional two-dimensional image detection problem into a target detection problem based on sequence data, and providing a new approach for real-time perception of hyperspectral targets in oil-bearing sea (river) areas. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is an analytical diagram of the original hyperspectral data acquired by the UAV in the method of this invention;

[0020] Figure 2 This is the overall design diagram for detecting non-imaging paradigm floating oil target areas in the method of this invention;

[0021] Figure 3 This diagram illustrates the detection process and results of the floating oil target area in the method of this invention. Detailed Implementation

[0022] To make the technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention:

[0023] like Figure 1 The method for detecting oil slicks using a UAV's hyperspectral non-imaging paradigm, as shown, specifically includes the following steps:

[0024] S1: First, use a drone equipped with a hyperspectral device to perform a parallel push-broom scan of the oil spill area at sea to directly obtain raw hyperspectral data of the measurable sea area.

[0025] S2: The hyperspectral vectors are seamlessly stitched together using a spectral mixing matrix pixel stitching method to obtain a hyperspectral pixel cube containing a large portion of clean sea area data, as well as non-oil-slick areas such as sea (river) coastlines and land.

[0026] S3: Using remote sensing detection and interpretation methods for oil slicks, the obtained hyperspectral pixel cubes are used as input data. One or more pixels are used as units to detect oil slicks using full-spectrum features. The resulting full-spectrum datasets of oil slicks and suspected oil slicks are used as the foreground, while retaining the relative positional information of the corresponding areas. Clean sea area data, coastlines, and land data (excluding oil slicks) are marked as background.

[0027] S31: Based on the existing hyperspectral data of the oil spill area, determine the full-spectral characteristic information of the oil spill area and other locations at the same time.

[0028] S32: Spectral comparison algorithms such as spectral angle, spectral feature matching and binary coding are used to compare the spectral segments of the region to be tested, and different detection weight parameter values ​​are set for each comparison method.

[0029] S33: The identification score is used to determine the oil spill area and suspected oil spill area, and the type of clean sea surface.

[0030] S34: Determine the threshold to obtain foreground data of oil slick areas and suspected oil slick areas, as well as background data of non-oil slick areas, including clean sea areas, coastlines and land.

[0031] S4: The background is represented by a hyperspectral cubic sparse representation, and sparse matrix transformation is used to compress the sparse rows of spectral bands. The feature dataset of the oil spill area and the sparse dataset obtained through inter-spectral degradation are used as input data for transmission, preparing for the transmission of hyperspectral data of the oil spill area to a ground remote sensing analysis platform. The acquisition of raw spectral data, the stitching of spectral mixing matrix data, and the remote sensing detection and interpretation methods for the oil spill area are all completed before the transmission of hyperspectral data of the oil spill area.

[0032] S41: Construct a sparse measurement matrix for random mapping, where the mapping conditions satisfy the constraint of isometry, effectively preserving the spectral information of the original full-spectrum data in non-floating oil areas. Specifically, the mapping satisfies:

[0033] y =R x Where x represents the high-dimensional data of each measured pixel, and its x-dimensionality is x∈R d Let y represent the low-dimensional data obtained by reconstruction through a random mapping matrix, y∈R. k The dimension of the corresponding random measurement mapping matrix R is k. d (k≤d), all elements r ij The probability density function is a ~N(0,1) distribution that satisfies:

[0034]

[0035] S42: To determine whether the sparse representation vector obtained by spectral band degradation is suitable, a spectral band selection affinity evaluation criterion is introduced to minimize the loss of original information caused by degradation. The affinity loss function is:

[0036]

[0037] Where X is the low-dimensional matrix we obtain through random mapping, D represents the low-dimensional combined feature spectrum, Z represents the sparse coefficient matrix, and Z is the iteration coefficient required for spectral matrix compression.

[0038] S43; By using the ADM optimization learning algorithm, a sparse dataset obtained through inter-spectral degradation is finally obtained, achieving the effect of foreground spectral degradation.

[0039] S5: Mark the oil slick area containing relative location information into the background image, and quickly detect and plot the oil slick area focusing results on the map, which to some extent solves the data redundancy problem of secondary processing of the original hyperspectral data of the oil slick area.

[0040] Example:

[0041] Description of oil spill simulation experiment scenario and basic principles of spectrometer camera

[0042] This study used indoor experimental methods to obtain hyperspectral reflectance data of marine oil spill slicks. The seawater samples used were from the Yellow Sea in China, collected under clear, windless weather conditions, free from the influence of seabed suspended organisms and sediment, thus reflecting the clarity and brightness typical of deep-sea nearshore waters. Portable polyvinyl chloride (PVC) buckets were used, made of a dark blue, water-insoluble plastic material to minimize alteration to the underlying color of the deep-sea water. The oils used in the indoor experiments included marine diesel (refined oil), marine fuel oil (heavy oil), and Brazilian crude oil. Parameters for the different oils are shown in Table 1 below.

[0043] Table 1. Detailed parameters of the oil samples used in the experiment.

[0044]

[0045] 1 American Petroleum Institute (API): Standards of the American Petroleum Institute

[0046] The preparation process for the oil film sample involved first using a magnetic levitation stirrer to heat and stir the oil sample simultaneously. A specific volume of oil sample was then dropped onto the surface of the experimental seawater using a specialized experimental oil sample tube. The experimental seawater was then injected into the experimental water tank. To obtain an oil film with a thickness of 1 μm, calculations were performed to determine the required oil film thickness per 1 m... 3 A 1cm drop of oil must be placed on the surface of the seawater sample. 3The oil sample was prepared by immersing the oil sample tube in the corresponding oil sample dish 3-5 times to reduce subjective bias in oil film thickness caused by oil adsorption in the measuring tube. Based on the calculated proportions, 225 μL of sample oil was added to the experimental container. Since oil density is lower than water density, it forms a floating oil layer on the water surface. The floating oil seawater sample was placed in a heated experimental oil sample container at a constant temperature (around 40℃) to allow the floating oil to diffuse evenly and form an oil film. The oil film experiment was conducted in a dark room under windless conditions to eliminate interference and clutter. A zenith angle arch support with a radius of 2.3 m was pre-built in the dark room to measure the incident and outgoing optical fibers. A pushbroom hyperspectral camera was mounted on top of the arch support, and a constant-speed motor inside the arch support drove the spectrometer to perform uniform sweeping of the water surface. See the attached document for details of the original hyperspectral data acquisition by the UAV. Figure 1 .

[0047] Comparison of raw experimental data and overall design diagram for detecting oil slick targets without imaging

[0048] The hyperspectral remote sensing image measured by the grating slit of a pushbroom hyperspectral camera is a spatial-spectral data cube. Each layer of the cube represents a band, and the attribute values ​​of each band for each pixel constitute a spectral vector, i.e., a spectral curve. Along the time dimension, the hyperspectral curves of different pixels are stitched together along the UAV's trajectory to obtain a continuous time-dimensional hyperspectral cube. This cube includes hyperspectral data of the sea surface, oil-water interface, and oil slicks. Each experiment lasts 10 seconds, with a frame rate of 10 frames per second, resulting in over 1900 spectral bands and over 1450 grating slit pixels. The final result is a mixed cube matrix of DN values, consisting of over 28 million values ​​measured every 10 seconds. We first label the oil slick and suspected oil slick areas on the sea surface as the foreground, and the clean sea area data, coastlines, and land (non-oil slick areas) as the background. The segmentation of oil slick and suspected oil slick areas utilizes an oil slick area detection and classification module. A deep learning-based image processing platform is used to train and register the oil slick foreground features onto the original mixed cubic matrix. Unregistered non-oil slick areas undergo sparse matrix processing, and an algorithm is used to sparsely compress non-oil slick feature regions, including seawater and oil-water boundaries. The detected oil slick areas are highlighted, and their images are simultaneously fed into the compressed storage module. The oil spill area information is transmitted to a ground-based remote sensing imaging analysis platform for imaging and target tiling of the oil spill area. The overall design diagram for non-imaging paradigm oil slick target area detection is attached. Figure 2 .

[0049] The result image obtained by detecting the target area of ​​the floating oil using raw measurement data is a hyperspectral image.

[0050] By using existing hyperspectral remote sensing data analysis platforms, the precise location of oil slicks can be displayed in the original images, and the focusing coordinates (latitude and longitude) of the oil slicks can be quickly plotted and displayed. At the same time, the amount of information that needs to be processed on the analysis platform is greatly reduced, transforming the traditional two-dimensional image detection problem into a target detection problem based on sequence data, and providing a new approach for real-time target perception in hyperspectral remote sensing of oil slicks.

[0051] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for detecting oil slicks using a hyperspectral non-imaging paradigm from an unmanned aerial vehicle (UAV), characterized in that... include: Parallel push-broom scanning was used to acquire raw hyperspectral data of the measured sea area in the oil spill zone. A spectral mixing matrix pixel stitching method is used to seamlessly stitch together hyperspectral raw data to obtain hyperspectral pixel cubes of clean sea areas, coastlines, and non-oil-slick areas on land. Using hyperspectral pixel cubes as input data, oil spill areas are detected based on the full spectrum features of oil spill, with one or more pixels as units. The full spectrum dataset of oil spill areas and suspected oil spill areas is used as the foreground, and the relative position information of the corresponding areas is preserved. Clean sea area data, coastline and land non-oil spill area data are marked as background. The background is represented by a hyperspectral cube sparse representation, and the sparse dataset is obtained by spectral band sparse row compression using matrix mapping. The feature dataset and sparse dataset of the oil spill area are used as the transmission data. Mark the oil slick area containing relative location information onto the background image, and plot the focused oil slick area on the map; When using matrix mapping to compress sparse rows of spectral segments to obtain sparse datasets: Construct a sparse measurement matrix and perform random mapping, where the mapping condition satisfies the constraint of isometry. Specifically, the mapping satisfies: y =R x Where x represents the high-dimensional data of each measured pixel, and its x-dimensionality is x∈R d Let y represent low-dimensional data obtained by reconstruction using a random mapping matrix, y∈R k The dimension of the corresponding random measurement mapping matrix R is k. d, k≤d, all elements r ij The probability density function is a ~N(0,1) distribution that satisfies: A spectral band selection affinity evaluation criterion is introduced, which uses an affinity loss function to determine whether the sparse representation vector obtained by spectral band degradation is suitable. Where X is a low-dimensional matrix obtained through random mapping, D represents the low-dimensional combined feature spectral band, and Z represents the sparse coefficient matrix, which is the iteration coefficient required for spectral matrix compression, thereby achieving the effect of foreground spectral band degradation.

Images