A method for monitoring fish habitat protection in waters based on multi-source satellite remote sensing images

Abstract

The invention discloses a method for protective monitoring of fish habitats in water areas based on multi-source satellite remote sensing images, comprising: comprehensively acquiring SAR and visible light satellite remote sensing images of multiple temporal and spatial scales and preprocessing images of different modalities respectively; then extracting the water area range based on the normalized difference water index (NDWI) of the visible light image and the polarized water index (SDWI) of the SAR image, and calculating the chlorophyll concentration parameter (YChla) and the suspended solids concentration parameter (YTSS) to evaluate the water quality; monitoring abnormal fishing and sand mining dynamics based on the YoloV3 target recognition model and combining the characteristics of vessels with abnormal behavior; extracting the changes in the water area and the surrounding areas of the fish habitat in different periods by using satellite remote sensing images, timely discovering engineering construction activities and lake reclamation behaviors, judging the impact on the fish habitat, and then guiding relevant personnel to deal with it.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing image processing for fishery protection, and in particular relates to a method for monitoring the protective performance of fish habitats in water areas based on multi-source satellite remote sensing images. Background Art

[0002] The main characteristics of fish habitats are their large spatial scale and wide distribution. Compared with the ocean and ordinary inland environment, the plateau ecosystem is more fragile and is more significantly affected by factors such as human activities and climatic conditions. The deterioration of the aquatic ecological environment, overfishing by humans and improper production activities are the main factors leading to the continuous reduction or even extinction of aquatic fish resources and endangered fish species in plateau areas. The role of large-scale, dynamic and continuous monitoring of these key factors in protecting aquatic fish habitats is self-evident.

[0003] In recent years, with the large-scale launch of remote sensing satellites carrying various sensors, satellite remote sensing technology has demonstrated its application value in the fields of rational development and utilization of resources, ecological environment protection, etc., with its advantages of wide spatial coverage, high revisit frequency, and rich sensor types. In terms of high-frequency dynamic monitoring of fishery resource protection and production activities, satellite remote sensing technology currently focuses more on the application of oceans and ordinary inland waters, and there is less research on large-scale natural waters in plateau areas. In view of this, there is an urgent need to provide a method for protecting fish habitats in inland plateau waters based on multi-source satellite remote sensing images, so as to promptly discover potential natural disasters and human violations in the process of fishery resource production and protection. Summary of the invention

[0004] Purpose of the invention: In view of the above problems, the present invention proposes a protective monitoring method for aquatic fish habitats based on multi-source satellite remote sensing images, which utilizes multi-temporal and spatial scale, multi-spectral comprehensive satellite remote sensing images to construct a protective monitoring method for aquatic fish habitats that integrates monitoring contents such as water quality, abnormal fishing, and improper human engineering activities, so as to provide technical support for fish habitat quality assessment, species protection, and rational utilization and sustainable development of fishery resources.

[0005] Technical solution: To achieve the purpose of the present invention, the technical solution adopted by the present invention is: a method for monitoring the protective habitat of fish in waters based on multi-source satellite remote sensing images, comprising the following steps:

[0006] First, we acquire SAR and visible light satellite remote sensing images at multiple temporal and spatial scales, including multispectral optical satellite remote sensing images with a spatial resolution of 0.5-30 meters and SAR images with a spatial resolution of 3-20 meters. The highest update frequency of temporal resolution is daily, and the lowest update frequency is monthly. We also pre-process images of different modalities separately.

[0007] Then, the water area is extracted based on the normalized water index of multispectral optical satellite remote sensing images and the polarized water index of SAR images, and the chlorophyll concentration and suspended matter concentration are calculated to evaluate the water quality, which is used to evaluate the underwater living environment of fish.

[0008] Then, based on the YoloV3 target recognition model, abnormal targets in the waters are extracted and identified, and abnormal fishing and sand mining dynamics are monitored in combination with the characteristics of abnormal ship behavior;

[0009] Finally, satellite remote sensing images are used to extract changes in the waters and surrounding areas of fish habitats at different times, to discover engineering construction activities and lake reclamation behaviors, to determine the impact on fish habitats, and then to guide relevant personnel in taking measures.

[0010] Furthermore, the water area extraction process is as follows:

[0011] Optical satellite remote sensing images are used to extract the range of inland lakes and river sections based on the normalized water index. The calculation formula is:

[0012] NDWI=(B _G -B _NIR ) / (B _G +B _NIR )

[0013] Among them, NDWI refers to the normalized water index calculated based on multispectral optical satellite remote sensing images. _G and B _NIR They represent the green band and near-infrared band of the multispectral image, respectively. 0.3 is used as the default threshold of NDWI. NDWI>0.3 indicates water body, and vice versa.

[0014] When optical images are unavailable, SAR images are used to extract the range of inland lakes and river sections based on the polarization water index. The calculation formula is:

[0015] SDWI=ln(10×B _VV ×B _VH )-10

[0016] Among them, SDWI refers to the polarization water index calculated based on SAR images, B _VV Represents the band in which SAR signals are transmitted and received vertically, B _VH The band represents vertical transmission and horizontal reception. 0.3 is used as the default threshold of SDWI. SDWI>0.3 indicates water body, otherwise it is not water body.

[0017] Furthermore, the empirical model of water chlorophyll a and suspended solids is used to invert the key parameters of water quality. The key parameters refer to the concentration of chlorophyll a and suspended solids in the water body, which are calculated as follows:

[0018] YChla=k*[B _RE -B _R ) / (B _RE +B _R ]+b

[0019] Among them, YChla refers to the chlorophyll a concentration value calculated based on multispectral satellite remote sensing images, B _RE and B _R are the red edge band and red band in the multispectral image, respectively; k and b are empirical coefficients, and the default values ​​of k and b are set to 15.5 and 5.6 respectively; the chlorophyll a concentration exceeding 30 μg / L is considered to be substandard water quality; the suspended matter concentration is calculated using the following formula:

[0020] YTSS=c×B _R 2 –d×B _R +e

[0021] Among them, YTSS refers to the suspended matter concentration value calculated based on multispectral satellite remote sensing images, B _R is the red band in the multispectral image, c, d, e are empirical coefficients, and the default values ​​are set to 0.4, -0.2, and 0.2 respectively; the suspended matter concentration exceeding 25 mg / L is considered to be substandard water quality.

[0022] Furthermore, the abnormal fishing and sand mining dynamic monitoring process is as follows:

[0023] First, the extracted water area is used as the monitoring target area, and optical satellite images with a resolution better than 3 meters are used to take photos every day;

[0024] Then, we built the YoloV3 target recognition model and used the DOTA remote sensing dataset containing various types of ship target images as training samples to train the YoloV3 model.

[0025] Then, the collected visible light satellite images are input into the trained YoloV3 model to identify the ship targets. When optical satellite images cannot be obtained, SAR images with a resolution better than 10 meters are used to compare the intensity of the previous and next phases:

[0026]

[0027] Among them, K_ratio represents the intensity ratio of the SAR images before and after the phase, K b,i and K a,i are the pixel values ​​in the water area of ​​the previous and next phase SAR images respectively, and n is the number of pixels in an image; when K_ratio is less than the threshold, it is determined that there is a ship target;

[0028] Whether the vessel target is an abnormal target is determined based on its external shape, the area where it appears and the time it stays there. The abnormal target is an abnormal fishing vessel or a sand mining vessel. The size, longitude and latitude and local screenshot-related features of the identified abnormal target are extracted.

[0029] Furthermore, the monitoring process of engineering construction activities and lake reclamation is as follows:

[0030] The impact zone is 10 kilometers outside the outer boundary of the water area, and abnormal man-made production and construction activities in the area are monitored and identified;

[0031] First, extract the outer boundary of the latest water area, and compare it with the previous boundary using single lakes and connected river sections as units. For abnormally changing shore sections, retrieve the optical satellite images before and after for comparison to check the actual changes of the ground features and confirm whether there is any destructive reclamation behavior.

[0032] Secondly, the maximum likelihood supervised classification method was used to classify the features in the impact area into grassland, forest land, construction land, agricultural land, water system, and road.

[0033] By comparing the differences between the latest classification results and the basic classification results, the mutual transfer between land feature types was identified. The basic classification results were obtained through the remote sensing interpretation algorithm of land feature classification. The impact of construction land on its surrounding water systems was evaluated based on the GIS overlay analysis method to determine the impact on downstream lake waters.

[0034] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial technical effects:

[0035] The present invention combines SAR and visible light multimodal images of multiple spatiotemporal scales, and provides data support for the protection of fish habitats from a more comprehensive, rapid and accurate perspective. The prior art mainly explains from one aspect of fish habitats, such as abnormal fishing, which lacks systematicity. The present invention monitors water quality and abnormal human illegal behaviors from two large fish habitat living spaces, above the water area and around the water area, highlighting the integrity of protection. The present invention introduces deep learning into the identification process of water targets. Compared with conventional visual viewing methods, it is more efficient and has better application prospects in the context of large-area water monitoring and remote sensing large data acquisition.

[0036] The present invention realizes the protective monitoring of fish habitats in waters based on multi-source satellite remote sensing images. In terms of the acquisition of comprehensive satellite images, at the county scale, optical images with a resolution of 3 meters and SAR images with a resolution of 5 meters are better than those at the county scale, which meet the full coverage requirements of 1-3 days; at the provincial scale, optical images with a resolution of 20 meters and SAR images with a resolution of 30 meters are better than those at the provincial scale, which meet the full coverage requirements of 3-10 days, and have the image guarantee requirements for dynamic monitoring of large-scale waters; in terms of monitoring abnormal fishing and sand mining behaviors, based on optical satellite images with a resolution of 2 meters, the recall rate of identifying ship targets using artificial intelligence algorithms is 0.85, and the accuracy rate is 0.8, which can meet the requirements for identifying and extracting abnormal targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0038] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0039] The present invention discloses a method for monitoring fish habitats in waters based on multi-source satellite remote sensing images. By comprehensively utilizing SAR and visible light multimodal images at multiple spatial scales, such as sub-meter, meter, and decameter, and multiple time scales, such as daily, monthly, and seasonal acquisitions, a system for protecting fish habitats in waters of a region is constructed to monitor water quality, abnormal fishing, and improper human engineering construction activities. The specific contents are as follows: (1) Acquisition and preprocessing of comprehensive satellite remote sensing images. According to the size of the water monitoring area, the source of satellite remote sensing images is determined. At the county level, optical images with a resolution better than 3 meters and SAR images with a resolution better than 5 meters are used. At the provincial level, considering the effective coverage of satellite images, 10-meter and 16-meter optical images and 8-meter, 10-meter, 20-meter, and 25-meter SAR satellite images are used to supplement the provincial level. Based on the acquired images, radiation correction, atmospheric correction, and orthorectification are performed on the optical images. (2) Water quality monitoring. First, the corrected images are used to extract the water area in the monitoring area. , and then calculate the chlorophyll concentration and suspended solids concentration for each lake and river section, and refer to the commonly used water chlorophyll and suspended solids concentration evaluation standards to divide the water body into five levels: excellent, good, general, poor and bad, and extract the location and spatial range related information of the last three types of water bodies; (3) Monitoring of abnormal fishing and sand mining behavior: Use artificial intelligence technology to extract and identify abnormal targets on the water area, and provide the size, location, and external morphology related information of the abnormal targets to professionals for further verification, and then determine the legality of the behavior; (4) Monitoring of improper human engineering activities: For the impact area within a certain range (such as 1 km) around the water area, use the supervised classification method to divide the land object types, and compare the latest period with the previous classification results to extract the transfer and change of the land object, and superimpose the extracted water system vectors on the newly added buildings and structures and the construction sites in the construction process for analysis to evaluate the impact of engineering activities on the surrounding fish habitats. The specific process is as follows: Figure 1 , the implementation includes the following steps:

[0040] ① Obtain and process comprehensive satellite remote sensing images; including multispectral optical satellite images with a spatial resolution of 0.5-30 meters and SAR images with a resolution of 3-20 meters; the maximum update frequency of temporal resolution is daily and the minimum update frequency is monthly; the satellite remote sensing images are geometrically and radiometrically corrected using the attached RPC parameters and publicly available 30-meter SRTM DEM data.

[0041] The reason for choosing 0.5-30m multispectral images is that the highest resolution of satellite remote sensing is about 0.3m, but the amount of data captured is small, and the general capture of remote sensing images is currently required to be at least 0.5m or lower, so the currently commonly used highest resolution is 0.5m. Considering the monitoring effect of fish habitats, the monitoring effect of more than 30m is generally not ideal, so multispectral images with a maximum resolution of 30m are selected. The reason for choosing 3-20m SAR images is closer to the above. 3m is the highest resolution of the most commonly used SAR images, and 20m is the boundary between medium resolution and low resolution.

[0042] ② Extraction of water area and inversion of key water quality parameters; First, the scope of inland lakes and river sections is extracted based on the normalized water index using optical satellite images:

[0043] NDWI=(B _G -B _NIR ) / (B _G +B _NIR )

[0044] Among them, NDWI refers to the normalized difference water index calculated based on multispectral optical satellite images. _G , B _NIR They represent the green band and near-infrared band of the multispectral image, respectively. 0.3 is used as the default threshold of NDWI. NDWI>0.3 indicates water body, and vice versa. When the optical image cannot be obtained due to the influence of cloud and rain, SAR image is used as a supplement to the acquisition of water area. The calculation formula is:

[0045] SDWI=ln(10×B _VV ×B _VH )-10

[0046] Among them, SDWI refers to the polarization water index calculated based on SAR images, B _VV Represents the band in which SAR signals are transmitted and received vertically; B _VH The band represents vertical transmission and horizontal reception. 0.3 is used as the default threshold of SDWI. SDWI>0.3 indicates water body, otherwise it is not water body.

[0047] Secondly, the empirical model of chlorophyll a and suspended solids in water body is used to invert the key parameters of water quality. The key parameters refer to the concentration of chlorophyll a and suspended solids in water body, which are calculated as follows:

[0048] YChla=k*[B _RE -B _R ) / (B _RE +B _R ]+b

[0049] Among them, YChla refers to the chlorophyll a concentration value calculated based on multispectral remote sensing images, B _RE and B _R They are the red edge band and red band in the multispectral image, respectively. k and b are empirical coefficients. These two values ​​are related to the characteristics of the water body in the area. The default values ​​used are set to 15.5 and 5.6 respectively. Chlorophyll a concentration exceeding 30 μg / L is considered to be substandard water quality. The suspended matter concentration is calculated using the following formula:

[0050] YTSS=c×B _R 2 –d×B _R +e

[0051] Among them, YTSS refers to the suspended matter concentration value calculated based on multispectral remote sensing images, B _R is the red band in the multispectral image, c, d, and e are empirical coefficients, and the default values ​​are set to 0.4, -0.2, and 0.2, respectively. Suspended matter concentration exceeding 25 mg / L is considered substandard water quality.

[0052] ③ Dynamic monitoring of abnormal fishing and sand mining; these two situations are identified based on the identified appearance characteristics of the ships. For example, compared with general abnormal fishing ships, sand mining ships are larger in length and scale, and are equipped with sand mining equipment, while inland fishing boats are smaller and have simpler structures.

[0053] First, the extracted water area is used as the monitoring target area, and optical satellite images with a resolution better than 3 meters are used to take key photos every day. Then, AI is used to automatically detect and identify water targets. The training samples use the public DOTA remote sensing data set (including images of various types of ship targets), and the network model uses YoloV3. The collected visible light satellite images are input into the trained model to identify various ship targets. In the case where high-resolution optical satellite images cannot be obtained, SAR images with a resolution better than 10 meters are used to compare the intensity of the before and after phases:

[0054]

[0055] Among them, K_ratio represents the intensity ratio of the two SAR images before and after. For example, a SAR image was obtained yesterday, and a SAR image of the same area was obtained today. The two SAR images are compared with the corresponding waters to obtain a ratio image. If there was no ship at a certain location in the waters yesterday, the image value would be small. If there is a ship today, the image value at the same location becomes larger, and there will be a difference between the two. n refers to the number of pixels in an image; after identifying the abnormal area, the same as the above process, the type of abnormal behavior is determined based on the size of the ship, external shape and other features. K b,i and K a,iThey are the pixel values ​​within the water area of ​​the SAR images of the previous and next phases, respectively, and the default comparison threshold is set to 0.8. When K_ratio is smaller, that is, the larger the value of the later phase image, the more likely it is to confirm the ship target. Therefore, when the threshold is set to be less than 0.8, it is considered that there may be a ship target. This threshold is determined by experience. After deep learning target recognition of visible light and 0.8 threshold extraction of SAR images, the ship target will be obtained. At this time, professionals are required to judge whether it is an abnormal target based on visible light and SAR images. The abnormal target is judged based on its external morphology, the area where it appears, and the residence time. The size, longitude and latitude, and local screenshots of the identified abnormal targets are extracted and handed over to professionals for verification.

[0056] ④ Monitoring of engineering construction activities and lake reclamation; The outer boundary of the water area is expanded by 10 kilometers as the impact area, and abnormal human production and construction activities in the area are monitored and identified. First, the outer boundary of the water area in the latest period is extracted, and a single lake and connected river section are used as units to compare with the previous period. For abnormally changed shore sections, high-resolution satellite images before and after are retrieved for comparison to check the actual changes of the features and confirm whether there is any reclamation and destruction; secondly, the features in the impact area are classified using the maximum likelihood supervised classification method, and divided into grassland, forest land, construction land, agricultural land, water system, road and other types of features. The difference between the latest classification results and the basic classification results (obtained through the remote sensing interpretation algorithm of feature classification, such as the support vector machine (SVM) supervised classification method) is compared to identify the mutual transfer between feature types, and the impact of construction land on its surrounding water system is evaluated based on the GIS overlay analysis method to determine the impact on the downstream lake waters.

Claims

1. A method for monitoring the protective habitat of fish in waters based on multi-source satellite remote sensing images, characterized in that: The following steps are involved: First, we acquire SAR and visible light satellite remote sensing images at multiple temporal and spatial scales, including multispectral optical satellite remote sensing images with a spatial resolution of 0.5-30 meters and SAR images with a spatial resolution of 3-20 meters. The highest update frequency of temporal resolution is daily, and the lowest update frequency is monthly. We also pre-process images of different modalities separately. Then, the water area is extracted based on the normalized water index of multispectral optical satellite remote sensing images and the polarized water index of SAR images, and the chlorophyll concentration and suspended matter concentration are calculated to evaluate the water quality, which is used to evaluate the underwater living environment of fish. Then, based on the YoloV3 target recognition model, abnormal targets in the waters are extracted and identified, and abnormal fishing and sand mining dynamics are monitored in combination with the characteristics of abnormal ship behavior; The dynamic monitoring process of abnormal fishing and sand mining is as follows: First, the extracted water area is used as the monitoring target area, and optical satellite images with a resolution better than 3 meters are used to take photos every day; Then, we built the YoloV3 target recognition model and used the DOTA remote sensing dataset containing various types of ship target images as training samples to train the YoloV3 model. Then, the collected visible light satellite images are input into the trained YoloV3 model to identify the ship targets. When optical satellite images cannot be obtained, SAR images with a resolution better than 10 meters are used to compare the intensity of the previous and next phases: Among them, K_ratio represents the intensity ratio of the SAR images before and after the phase, K b,i and K a,i are the pixel values ​​in the water area of ​​the previous and next phase SAR images respectively, and n is the number of pixels in an image; when K_ratio is less than the threshold, it is determined that there is a ship target; Determine whether the vessel target is an abnormal target based on its external shape, the area where it appears, and the time it stays there. The abnormal target is an abnormal fishing vessel or a sand mining vessel. Extract the size, longitude and latitude, and local screenshot-related features of the identified abnormal target. Finally, satellite remote sensing images are used to extract changes in the waters and surrounding areas of fish habitats at different times, to discover engineering construction activities and lake reclamation behaviors, to determine the impact on fish habitats, and then to guide relevant personnel in taking measures.

2. The method for monitoring fish habitat protection in waters based on multi-source satellite remote sensing images according to claim 1, characterized in that: The water area extraction process is as follows: Optical satellite remote sensing images are used to extract the range of inland lakes and river sections based on the normalized water index. The calculation formula is: NDWI=(B _G -B _NIR ) / (B _G +B _NIR ) Among them, NDWI refers to the normalized water index calculated based on multispectral optical satellite remote sensing images. _G and B _NIR They represent the green band and near-infrared band of the multispectral image, respectively. 0.3 is used as the default threshold of NDWI. NDWI>0.3 indicates water body. On the contrary, it is not a body of water; When optical images are unavailable, SAR images are used to extract the range of inland lakes and river sections based on the polarization water index. The calculation formula is: SDWI=ln(10×B _VV ×B _VH )-10 Among them, SDWI refers to the polarization water index calculated based on SAR images, B _VV Represents the band in which SAR signals are transmitted and received vertically, B _VH The band represents vertical transmission and horizontal reception. 0.3 is used as the default threshold of SDWI. SDWI>0.3 indicates water body, otherwise it is not water body.

3. The method for monitoring fish habitat protection in waters based on multi-source satellite remote sensing images according to claim 1 is characterized in that: The key water quality parameters are inverted using the empirical model of water chlorophyll a and suspended solids. The key parameters refer to the concentration of chlorophyll a and suspended solids in the water body, which are calculated as follows: YChla=k*[B _RE -B _R ) / (B _RE +B _R ]+b Among them, YChla refers to the chlorophyll a concentration value calculated based on multispectral satellite remote sensing images, B _RE and B _R are the red edge band and red band in the multispectral image, respectively; k and b are empirical coefficients, and the default values ​​of k and b are set to 15.5 and 5.6 respectively; the chlorophyll a concentration exceeding 30 μg / L is considered to be substandard water quality; the suspended matter concentration is calculated using the following formula: YTSS=c×B _R 2 –d×B _R +and Among them, YTSS refers to the suspended matter concentration value calculated based on multispectral satellite remote sensing images, B _R is the red band in the multispectral image, c, d, e are empirical coefficients, and the default values ​​are set to 0.4, -0.2, and 0.2 respectively; the suspended matter concentration exceeding 25 mg / L is considered to be substandard water quality.

4. The method for monitoring fish habitat protection in waters based on multi-source satellite remote sensing images according to any one of claims 1 to 3, characterized in that: The monitoring process of engineering construction activities and lake reclamation is as follows: The impact zone is 10 kilometers outside the outer boundary of the water area, and abnormal man-made production and construction activities in the area are monitored and identified; First, extract the outer boundary of the latest water area, and compare it with the previous boundary using single lakes and connected river sections as units. For abnormally changing shore sections, retrieve the optical satellite images before and after for comparison to check the actual changes of the ground features and confirm whether there is any destructive reclamation behavior. Secondly, the maximum likelihood supervised classification method was used to classify the features in the impact area into grassland, forest land, construction land, agricultural land, water system, and road. By comparing the differences between the latest classification results and the basic classification results, the mutual transfer between land feature types was identified. The basic classification results were obtained through the remote sensing interpretation algorithm of land feature classification. The impact of construction land on its surrounding water systems was evaluated based on the GIS overlay analysis method to determine the impact on downstream lake waters.

Images