Intelligent auxiliary system and method for evaluating damage of coral reef

Abstract

The invention relates to the field of damage assessment, in particular to an intelligent auxiliary system and method for coral reef damage assessment, and the system comprises an image collection module which is used for collecting coral reef images through an underwater camera; the recognition module comprises a trained convolutional neural network, and the convolutional neural network is used for inputting the coral reef image and outputting the proportion, coverage rate, area and death rate of each coral reef variety; the density detection module is used for measuring the density of the collected coral reef samples, identifying the coral reef variety based on the identification module during measurement, and calculating the average density of a plurality of samples under the fixed coral reef variety; and the estimation module is used for inputting the damaged thickness evaluated on site, calculating the weight of the damaged coral reef based on the area and density of the coral reef, calculating the coral value of each variety based on a formula, and accumulating the coral values of each variety to obtain an economic loss value for evaluating the damage of the coral reef. By adopting the technical scheme of the invention, the manual work intensity and professional requirements of personnel in the evaluation process can be reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of damage assessment, in particular to an intelligent auxiliary system and method for coral reef damage assessment. BACKGROUND

[0002] Coral reefs are structures formed by animals of the order Scleractinia, which can be so large as to affect the physical and ecological conditions of their surroundings. Coral reefs exist in deep and shallow waters and are formed from the skeletons of thousands of tiny coral polyps made of calcium carbonate over hundreds to thousands of years. Coral reefs provide a habitat for many plants and animals, including worms, mollusks, sponges, echinoderms and crustaceans, and are also the nurseries of young fish in the ocean belt.

[0003] With the increasing frequency of human activities in the ocean, events such as pollution leaks, grounding and the like can have irreversible effects on coral reefs. The existing technology relies on manual investigation and analysis to determine the damage to coral reefs, which is labor-intensive and requires high technical expertise. Therefore, an intelligent auxiliary system and method for coral reef damage assessment that can reduce the labor intensity and expertise of personnel is needed. SUMMARY

[0004] To solve the above problems, the present application provides an intelligent auxiliary system and method for coral reef damage assessment, which can reduce the labor intensity and expertise of personnel in the evaluation process.

[0005] To achieve the above purpose, the technical solution of the present application is as follows: an intelligent auxiliary system for coral reef damage assessment, comprising:

[0006] An imaging acquisition module for acquiring coral reef images through an underwater camera;

[0007] An identification module comprising a trained convolutional neural network, the convolutional neural network being trained based on the species of coral and the live or dead state of the coral, and the convolutional neural network being used to input the coral reef images and output the proportion, coverage, area and mortality rate of each species of coral reef;

[0008] A density detection module for measuring the density of the collected coral reef samples, the measurement being based on the identification of the species of coral reef by the identification module, and the average density of multiple samples of a fixed species of coral reef being calculated;

[0009] An estimation module for inputting the damaged thickness of the on-site evaluation, calculating the weight of the damaged coral reef based on the area and density of the coral reef, and calculating the value of each species of coral based on the formula:

[0010] VAL = M * P * K1 * K2,

[0011] Wherein, VAL is the value of damaged coral reef, M is the weight of damaged coral reef, P is the benchmark price of coral, K1 is the protection value coefficient of coral, and K2 is the rare value coefficient of coral (usually 1, and when the actual market transaction price is greater than the evaluation price, the coefficient is the actual market transaction price divided by the evaluation price) ;

[0012] The values of various types of corals are added to obtain the economic loss value for evaluating the damage of the coral reef.

[0013] The above scheme has the following beneficial effects:

[0014] 1. In the scheme, the coral imaging of the damaged area is obtained through the underwater camera, and the trained convolutional neural network is used to assist the user in evaluating the damage of the coral according to the image. The convolutional neural network can determine the proportion, area, coverage rate and mortality rate of each type of coral in the image, so as to facilitate the evaluation of the damage of the coral reef caused by the damage event.

[0015] 2. In the scheme, the image analysis can only obtain the damage area, but cannot obtain the specific damage amount. Therefore, the density of the coral sample is detected. Based on the damage thickness caused by the damage event, the damage volume is calculated according to the damage area and damage thickness, and the total mass is calculated in combination with the density.

[0016] Further, the identification module is also used to distinguish soft corals, reef-building corals and reefs, and the density detection module and the estimation module are used for independent density detection and estimation of soft corals, reef-building corals and reefs.

[0017] Beneficial effect: The coral reef can be further subdivided, such as soft corals, reef-building corals and reefs with soft body, and separate density detection and estimation are performed according to the different properties.

[0018] Further, the density detection module obtains the weight of the coral reef sample based on weighing, and calculates the volume of the irregular coral reef sample by the drainage method.

[0019] Beneficial effect: The weight detection of the coral by the density detection module can be directly weighed, and the coral is mostly irregular, so the volume is detected by the drainage method, and the density of the coral is calculated in combination with the weight and the volume.

[0020] Further, in the density detection module:

[0021] The density of the reef-building coral = weight / drainage volume, and the volume = height * area * effective proportion coefficient, wherein the effective proportion coefficient = drainage volume / maximum length-width-height volume of the coral group, in order to quickly calculate, the effective proportion coefficient of the block-shaped reef-building coral group is 0.8, the effective proportion coefficient of the skin-shaped reef-building coral group is 0.6, the effective proportion coefficient of the leaf-shaped reef-building coral group is 0.4, and the effective proportion coefficient of the branch-shaped reef-building coral group is 0.2;

[0022] The soft coral density = weight / displacement volume, volume = height * area * effective proportionality coefficient, wherein the effective proportionality coefficient = displacement volume / maximum length-width-height volume of the coral colony, in order to quickly calculate, for different morphologies of the soft coral colony, the effective proportionality coefficient of the table cover shape is 0.8, the disc shape is 0.6, the flower shape is 0.4, the branch shape is 0.2, and the fan shape is 0.1.

[0023] Beneficial effects: the soft coral species are few, and the density is similar to water. Since the shape volume of the soft coral is unstable, the default value is directly used for density determination. Since the soft coral contains a large amount of water, but the sold coral is in a dry state, the volume of the soft coral is multiplied by 0.1 to fit the dehydrated volume change process of the soft coral.

[0024] Further, the value formula of the reef is calculated as reef weight * reef reference value * value damage coefficient, wherein the value damage coefficient is reef value coefficient - reef fragment value coefficient.

[0025] Beneficial effects: the reef is different from the coral, the coral will die, but the reef is only broken, and the broken reef still has utilization value, so the reef needs to be multiplied by a value coefficient for evaluating the change from the complete state to the damaged state.

[0026] Further, the convolutional neural network is also trained based on the coral reef damage reason, the proportion of each variety of the coral reef, the coral reef mortality rate, the coral reef shape, and the evaluated damaged thickness, and the convolutional neural network is used to input the coral reef damage reason and output the damaged thickness based on the coral reef image.

[0027] Beneficial effects: the image lacks stereoscopic information, so it is difficult to determine the damaged thickness relying on the depth information in the image. The coral reef can be damaged in different forms by ship grounding, collision, geological disasters, etc., and the damage situation caused by the same type of damage reason is similar, and for example, the ship displacement, collision strength, and geological disaster level can determine the damage intensity, so as to deduce the damaged thickness combined with the damage situation in the image.

[0028] Further, the convolutional neural network is also trained based on the undamaged coral reef image, the undamaged coral reef reef fragment image, the damage reason, and the damaged coral reef reef fragment image, and the convolutional neural network is used to input the damage reason and output the newly added amount of coral reef reef fragments caused by the damage reason based on the judgment of the coral reef image.

[0029] Beneficial effects: Reef fragments may exist before the coral reef is damaged, so in order to filter out the influence of the original fragments, the generation of fragments after damage needs to be evaluated. The convolutional neural network is trained on the images of undamaged coral reefs, undamaged coral reef fragments, damage causes, and damaged coral reef fragments, so as to master the amount of conventional fragments, the amount of conventional fragments generated by different damage causes, evaluate and determine the amount of newly added fragments.

[0030] Further, the imaging acquisition module is also used for denoising the coral reef image.

[0031] Beneficial effects: Denoising can remove burrs in the image and improve image clarity.

[0032] Further, the estimation module is also used for calculating the values of damaged reef-building coral, damaged soft coral, and damaged coral reef stones respectively, and adding the above three values to obtain the total economic loss value of the damaged coral reef.

[0033] Further, the estimation module is also used for calculating the ecological service function loss compensation fee, which is calculated based on the area of the damaged coral reef, the annual service value of the coral reef ecosystem per unit area, and the recovery period of the damaged coral reef.

[0034] Further, the estimation module is also used for calculating the resource damage repair fee, which includes the repair fee of damaged reef-building coral, the repair fee of damaged soft coral, and the repair fee of damaged coral reef stones; wherein the repair fee of damaged reef-building coral and damaged soft coral is calculated based on their respective damaged values by a certain multiple, and the repair fee of damaged coral reef stones is determined based on its damaged value and combined with expert recommendations.

[0035] An intelligent auxiliary method for coral reef damage assessment, comprising:

[0036] Step one, collect the coral reef images for training, which includes collecting undamaged coral reef images and damaged coral reef images in each sea area by using underwater cameras, and obtaining the damage causes, labeling the coral reef images and training the convolutional neural network;

[0037] Step two, determine the impact range according to the damage event, ride a waterborne vehicle to the damage event, release an underwater camera, collect underwater coral reef images within the impact range, and salvage or collect coral reef samples;

[0038] Step three, input the coral reef image into the convolutional neural network, and identify the density of the coral reef sample, combine the density information and the output result of the convolutional neural network, and calculate the loss of soft coral, reef-building coral and reef value respectively, add the values of soft coral, reef-building coral and reef to obtain the economic loss value for evaluating the damage of coral reef.

[0039] Beneficial effects: the convolutional neural network replaces the underwater area analysis of artificial evaluation of coral reef damage, reduces the work intensity and professional degree requirement, and also reduces the proportion of underwater work, simplifies the work flow.

[0040] Further, in step two, when the underwater camera collects images, a 50m-100m scale ruler is used to lay out on the collected section, and an underwater digital camera is used to shoot from one end of the ruler along the scale ruler.

[0041] Beneficial effects: the scale ruler can express length information, so that the length, width and height of the coral can be analyzed according to the image, and thus the area or thickness of the coral can be obtained.

[0042] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter in the description. BRIEF DESCRIPTION OF DRAWINGS

[0043] Fig. 1 The module schematic diagram of the intelligent auxiliary system for coral reef damage evaluation of the present application is shown in the figure.

[0044] Fig. 2 The schematic diagram of the logic embodiment of the intelligent auxiliary system for coral reef damage evaluation of the present application is shown in the figure.

[0045] Fig. 3 The step schematic diagram of the intelligent auxiliary method embodiment for coral reef damage evaluation of the present application is shown in the figure. DETAILED DESCRIPTION

[0046] The technical solutions of the present application will be described clearly and completely in combination with the accompanying drawings, obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the present application.

[0047] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0048] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0049] The following detailed description illustrates the specific implementation method:

[0050] Example 1:

[0051] As attached Figs. 1-3 As shown: An intelligent auxiliary system for assessing coral reef damage, comprising:

[0052] Imaging acquisition module: Used to acquire images of coral reefs through an underwater camera. The acquisition method can be monocular, binocular, or multi-view imaging. The imaging acquisition module is also used to denoise coral reef images.

[0053] The recognition module includes a trained convolutional neural network (CNN). The CNN is trained based on the coral species and the coral's life or death status. The CNN is used to take coral reef images as input and output the proportion, coverage, area, and mortality rate of each coral species. The CNN is also used to distinguish between soft corals, reef-building corals, and reefs. In the density detection and estimation modules, soft corals, reef-building corals, and reefs are all subjected to independent density detection and estimation.

[0054] Density detection module: used for measuring the density of the collected coral reef samples, the density is calculated based on the weight and volume of the coral reef, wherein the weight is based on the weighing of the coral reef, and the volume is measured by the displacement method, the measurement is based on the identification of the coral reef species by the identification module, and the average density of multiple samples of the fixed coral reef species is calculated; in the density detection module: the density of the reef-building coral = weight / displacement volume, volume = height*area*effective proportion coefficient, wherein the effective proportion coefficient = displacement volume / coral colony maximum length-width-height volume, in order to calculate quickly, for different morphologies of reef-building coral colony, the effective proportion coefficient of block type is 0.8, the effective proportion coefficient of skin shell type is 0.6, the effective proportion coefficient of leaf type is 0.4, and the effective proportion coefficient of branch type is 0.2;

[0055] The density of soft coral = weight / displacement volume, volume = height*area*effective proportion coefficient, wherein the effective proportion coefficient = displacement volume / coral colony maximum length-width-height volume, in order to calculate quickly, for different morphologies of soft coral colony, the effective proportion coefficient of block type is 0.7, the effective proportion coefficient of skin shell type is 0.6, the effective proportion coefficient of leaf type is 0.5, and the effective proportion coefficient of branch type is 0.3.

[0056] Estimation module: used for inputting the damaged thickness of the field evaluation, calculating the damaged coral reef weight based on the coral reef area and density, and calculating the value of each species of soft coral and reef-building coral based on the formula:

[0057] VAL = M*P*K1*K2,

[0058] wherein VAL is the value of the damaged coral reef, M is the weight of the damaged coral reef, P is the reference price of the coral species, K1 is the protection value coefficient of the coral, and K2 is the rare value coefficient of the coral (usually taken as 1, when the actual market transaction price is greater than the evaluation price, the coefficient is the actual market transaction price divided by the evaluation price);

[0059] The value formula of the reef stone is calculated as reef stone weight*reef stone reference value*value damage coefficient, wherein the value damage coefficient is the reef stone value coefficient-reef stone fragment value coefficient.

[0060] The values of each species of coral are added to obtain the economic loss value for evaluating the damage of the coral reef.

[0061] When used, the user needs to go to the damaged event to collect images and extract samples, and the damaged event is taken as the ship hull touching the coral reef and causing damage to the coral reef.

[0062] The user uses the imaging collection module to detect the imaging of the damaged seabed, and inputs the seabed imaging into the convolutional neural network, which can replace manual analysis to identify the sand body, reef stone, coral and coral species in the image, thereby saving the process of manual diving screening, and the operation personnel do not need to have relevant knowledge of coral classification.

[0063] Coral reefs are classified into soft corals, reef-building corals and reefs. Soft corals have high water content and soft texture, so it is necessary to overcome the problems of volume and density evaluation caused by water content and texture. In terms of weight and value calculation of soft corals, according to the unique physical properties of soft corals, the density determination follows similar principles as reef-building corals: density = weight / displacement volume, volume = height * area * effective proportionality coefficient, where effective proportionality coefficient = displacement volume / maximum length-width-height volume of coral colony. In order to achieve fast and accurate calculation, according to the different morphological groups of soft corals, some empirical values of effective proportionality coefficient are taken, such as 0.8 for table cover shape, 0.6 for disc shape, 0.4 for flower shape, 0.2 for branch shape, and 0.1 for fan shape. On this basis, considering the high water content and soft texture of soft corals, the volume change is significant and the law is difficult to measure during dehydration process, through a large amount of experimental data statistical analysis and actual case verification, the default density value 1000 Kg / m 3 is used to approximate the density of soft corals after dehydration, and the volume change ratio of soft corals after dehydration is set to 0.1. Therefore, when calculating the value of soft corals, the mass of underwater soft corals is the area of underwater soft corals * evaluation thickness * 1000 Kg / m 3 * 0.1, and then according to the relevant parameters such as the protection value coefficient and the rare value coefficient of soft corals, the value calculation formula specially designed for soft corals is substituted to accurately calculate the economic loss value of damaged soft corals.

[0064] For reefs, due to its nature as a collection of minerals, it has completely different characteristics from biological corals. Although the overall structure of the reef is destroyed after being broken, it still has a certain value according to its role in marine landscape shaping and coastal protection. In the evaluation of the value of damaged reefs, first, the density of the reef is measured by the density detection module, and then the volume of the damaged reef is calculated by combining the on-site evaluation data such as damaged thickness and area. Then, according to the reference value of the reef and the reef value coefficient and reef fragment value coefficient determined by comprehensively considering the influence of the reef before and after being broken on marine ecological environment and coastal stability, the value coefficient of the damage loss is calculated, and finally the reef value calculation formula is substituted to accurately quantify the economic loss corresponding to the damage of the reef.

[0065] Taking a specific coral reef damage event as an example, after a comprehensive and detailed on-site investigation and data collection of the damaged coral reef, the following key information is obtained: through professional measurement and accurate positioning, it is determined that the area of the coral damaged region is 160m 2The reefing coral in the area was damaged to varying degrees due to the sudden accident of the ship running aground, with the loss of coverage of branch-shaped reefing coral reaching 1.21%, and the loss of coverage of block-shaped reefing coral reaching 2.17%. At the same time, through repeated estimation by professional monitoring equipment and experienced technical personnel on site, it is concluded that the average thickness of the damaged reefing coral is about 0.1m.

[0066] In calculating the damaged mass of the reefing coral, the density calculation principle mentioned above is strictly followed, i.e. for reefing coral, density = weight / displacement volume, volume = height * area * effective proportionality coefficient (here, the effective proportionality coefficient for different forms of reefing coral is taken as: block 0.8, skin 0.6, leaf 0.4, branch 0.2), combined with the specific coral volume coefficient in the area. Through rigorous mathematical derivation and data processing, it is finally determined that the damaged branch-shaped reefing coral mass is 75.3Kg, and the block-shaped reefing coral weight reaches 304.3Kg.

[0067] Similarly, in this damaged area, soft coral also could not escape. Through on-site accurate monitoring, the loss of coverage of soft coral is 0.62%, and the estimated value of its average thickness of damage is about 0.05m. In the quality calculation of soft coral, according to the unique physical properties of soft coral and the corresponding density and volume calculation rules: density = weight / displacement volume, volume = height * area * effective proportionality coefficient (for different forms of soft coral, the effective proportionality coefficient is taken as: table cover 0.8, disc 0.6, flower 0.4, branch 0.2, fan 0.1), and considering the volume change law of soft coral during dehydration, the default density value 1000Kg / m 3 and the volume change ratio of soft coral after dehydration is set as 0.1 for approximate calculation. After a series of accurate operations, it is concluded that the damaged soft coral weight is 4.96Kg.

[0068] Looking at the reef part, the damaged reef area is determined to be 1000m 2 The impact caused by the grounding of the ship caused the average coverage of the reef in the area to decrease sharply to 72.75%, and the on-site monitoring estimated that the average thickness of the damage was about 0.1m. After calculating the density of the reef according to the reef density detection process, combined with the above data, it is concluded that the damaged reef mass is 124531.5Kg.

[0069] Further evaluation of the economic losses of various types of coral reefs, the commonly used various types of coral benchmark value on the market is 500 ￥ / Kg, and based on the consideration of the ecological protection significance of coral reefs, the protection coefficient is set to 5, and the rare coefficient is 1. For reef-building corals, according to the damaged mass and the corresponding value coefficient, a specially designed value calculation formula VAL = M*P*K1*K2 (where VAL is the value of damaged coral reefs, M is the weight of damaged coral reefs, P is the benchmark price of coral species, K1 is the protection value coefficient of coral, K2 is the rare value coefficient of coral) is used for calculation: the loss value of damaged branched reef-building corals is 75.3 Kg*500 ￥ / Kg*5*1 = 188250 ￥, the loss value of massive reef-building corals is 303.3 Kg*500 ￥ / Kg*5*1 = 760750 ￥, and the total value of damaged reef-building corals is 188250 ￥+760750 ￥=949000 ￥.

[0070] The total loss value of soft corals is also calculated according to the above value formula combined with the damaged mass, which is 4.96 kg*500 ￥ / kg*5*1 = 12400.0 ￥.

[0071] For reefs, since they still have a certain value after being broken, according to the difference between the reef value coefficient and the reef fragment value coefficient determined in the previous comprehensive evaluation (the difference is 0.1 after evaluation), the reef value calculation formula: reef value = reef weight*reef benchmark value*value damage coefficient, the total loss value of reef is 124531.5 kg*500 ￥ / kg*0.1 = 6226575 ￥.

[0072] Finally, the loss values of soft corals, reef-building corals and reefs, the three components of coral reefs, are added, that is, 949000 ￥+12400.0 ￥+6226575 ￥=7187975 ￥, so the total economic loss value caused by this coral reef damage event is accurately obtained, which provides solid and reliable data support for subsequent ecological restoration, responsibility identification and insurance claims.

[0073] Example 2:

[0074] The difference between the above embodiment and the present embodiment is that the convolutional neural network is also trained based on the coral reef damage reason, the proportion of each variety of coral reef, the coral reef mortality rate, the coral reef shape and the evaluated damaged thickness, and the convolutional neural network is used to input the coral reef damage reason and output the damaged thickness based on the coral reef image.

[0075] The image lacks stereo information, so it is difficult to determine the damaged thickness relying on the depth information within the image. The coral reef can be damaged in different forms by ship grounding, collision, geological disasters, etc., and the damage situation caused by the same type of damage reason is similar, while the damage intensity can be determined by, for example, the ship displacement, the collision strength, the geological disaster level, etc., so as to deduce the damaged thickness combined with the damage situation within the image. In this way, the number of times of the diver's work can be further reduced.

[0076] Embodiment 3:

[0077] The difference from the above embodiments is that the convolutional neural network is further trained based on the undamaged coral reef image, the undamaged coral reef reef fragment image, the damage reason, and the damaged coral reef reef fragment image, the convolutional neural network is used to input the damage reason, and based on the judgment of the coral reef image, the newly added amount of the coral reef reef fragment caused by the damage reason is output.

[0078] The reef fragment can exist before the coral reef is damaged, so in order to filter out the influence of the original gravel, the generation of the gravel after the damage needs to be evaluated. The convolutional neural network is trained based on the undamaged coral reef image, the undamaged coral reef reef fragment image, the damage reason, and the damaged coral reef reef fragment image, so as to master the amount of conventional original gravel, the amount of conventional gravel generated by different damage reasons, and evaluate and determine the newly added amount of the reef fragment.

[0079] More specifically, in order to more accurately and quickly determine the damage reason from the coral reef image and output the newly added amount of the coral reef reef fragment caused by the damage reason when the convolutional neural network processes the coral reef damage evaluation problem, the convolutional neural network needs to be deeply trained based on the undamaged coral reef image, the undamaged coral reef reef fragment image, the damage reason, and the damaged coral reef reef fragment image.

[0080] Before training the convolutional neural network, the input image data needs to be preprocessed and feature extracted. Let the input undamaged coral reef image be Iu, the undamaged coral reef reef fragment image be Ius, and the damaged coral reef reef fragment image be Ids.

[0081] Firstly, the image is normalized to map the image pixel value to the [0, 1] interval:

[0082]

[0083] where I represents the original image, represents the normalized image. Then, the feature extractor (such as SIFT, SURF, etc.) is used to extract the feature vector of the image. Let the extracted feature vector be f, then:

[0084]

[0085] To train the CNN, a proper loss function needs to be defined. Here we adopt a cross-entropy loss function combined with a mean squared error loss function to consider both classification and regression tasks. Let y be the true label of the damage cause (represented by one-hot encoding), the predicted probability distribution of the damage cause output by the CNN; z be the true reef fragment addition, the predicted value of the reef fragment addition output by the CNN. Then the loss function L is defined as:

[0086]

[0087] where α ∈ [0, 1] is a weight coefficient to balance the classification loss and the regression loss.

[0088] represents the cross-entropy loss function, and the calculation formula is:

[0089]

[0090] represents the mean squared error loss function, and the calculation formula is:

[0091]

[0092] where m is the number of damage cause categories, and m is the number of samples.

[0093] To improve the attention of the CNN to key features, an attention mechanism is introduced. Let the input feature map be where H, W, and C represent the height, width, and channel number of the feature map, respectively.

[0094] The attention weight is obtained through global average pooling and a fully connected layer:

[0095] s = Sigmoid(FC(GlobalAveragePooling(F)))

[0096] where Sigmoid is the activation function, FC represents the fully connected layer, and GlobalAveragePooling represents the global average pooling operation. Then, the attention weight is multiplied with the feature map to obtain the weighted feature map:

[0097] F' = F☉s

[0098] where ⊙ represents element-wise multiplication.

[0099] To capture feature information of different scales, a multi-scale feature fusion method is adopted. Let the feature maps output by different layers of the CNN be F1, F2,..., Fn, k, respectively, correspond to different scales. The feature maps of different scales are up-sampled or down-sampled to have the same size, and then the fused feature maps are spliced.

[0100] In order to further improve the accuracy of the damaged reason judgment, Bayesian inference is introduced. Let C represent the damaged reason category, and I represent the input coral reef image. According to Bayes' theorem, the posterior probability P(C|I) can be expressed as:

[0101]

[0102] Where P(I|C) is the likelihood probability, indicating the probability of observing the image I given the damaged reason C; P(C) is the prior probability, indicating the probability of the occurrence of the damaged reason C; P(I) is the evidence probability, indicating the probability of observing the image I. Since P(I) is the same for all damaged reason categories, it can be ignored. Therefore, only the relative size of P(I|C)P(C) needs to be calculated to determine the most likely damaged reason.

[0103] When outputting the new amount of reef fragments, a regression analysis method is used. Let the input feature vector be x, and the output new amount of reef fragments be z. Assume that the regression model is a linear regression model:

[0104] z=w T x+b

[0105] Where w is the weight vector and b is the bias term. In order to optimize the parameters of the regression model, the least squares method is used for training. The loss function is:

[0106]

[0107] The values of w and b are continuously updated by gradient descent method to minimize the loss function.

[0108] In the training process, stochastic gradient descent (SGD) or its variants (such as Adam, Adagrad, etc.) are used to minimize the loss function L. At each iteration, a small batch of samples is randomly selected from the training data set for training, and the parameters of the convolutional neural network are updated. In the inference process, the input coral reef image is preprocessed and feature extracted, and then input into the trained convolutional neural network. The network first outputs the predicted probability distribution of the damaged reason, and selects the class with the highest probability as the predicted damaged reason; at the same time, the network outputs the predicted value of the new amount of reef fragments.

[0109] The embodiment can improve the accuracy and speed of judging the damaged reason from the coral reef image, and more accurately output the new amount of reef fragments caused by the damaged reason.

[0110] An intelligent auxiliary method for coral reef damage assessment, comprising:

[0111] Step one, collect coral reef images for training, including collecting undamaged coral reef images and damaged coral reef images in each sea area using an underwater camera, and obtaining the damage causes, labeling the coral reef images and training a convolutional neural network;

[0112] Step two, determine the impact range according to the damage event, ride a watercraft to the damage event, and release an underwater camera to collect underwater coral reef images within the impact range, and salvage or collect coral reef samples;

[0113] Step three, input the coral reef images into the convolutional neural network, and identify and detect the density of the coral reef samples, combine the density information and the convolutional neural network output results, and calculate the loss of soft coral, reef-building stone coral and reef value respectively, and add the values of soft coral, reef-building stone coral and reef to obtain the economic loss value for assessing the damage of coral reef.

[0114] Beneficial effects: the convolutional neural network replaces manual underwater area analysis when assessing coral reef damage, reduces the work intensity and professional degree requirements, and also reduces the proportion of underwater work, simplifying the work process.

[0115] In step two, when the underwater camera collects images, a 50m-100m scale ruler is used to lay out on the cross section, and an underwater digital camera is used to shoot along the ruler from one end of the ruler.

[0116] The scale ruler can express length information, so that the length, width and height of the coral can be analyzed according to the image to obtain the area or thickness.

[0117] Embodiment 4

[0118] In a certain accident, the relevant weight calculation has been completed, and the values of reef-building stone coral, soft coral and coral reef stone are calculated respectively.

[0119] The loss calculation logic of the value of reef-building stone coral is as follows: the reference price of reef-building stone coral is 500 yuan / Kg, the species protection coefficient of reef-building stone coral is 5, and the case coefficient of reef-building stone coral is 1.0, so the damage calculation of the value of reef-building stone coral in the damaged area is as follows:

[0120] Table 1 Damage calculation of the value of reef-building stone coral in the damaged area

[0121]

[0122] The value of damaged reef-building coral = the weight of damaged reef-building coral * the reference value of coral * the protection coefficient of reef-building coral * the involvement coefficient of reef-building coral = 129.76 kg * 500 yuan / kg * 5 * 1 = 324400.00 yuan

[0123] The value of damaged reef-building coral is calculated to be about 324400.00 yuan.

[0124] The value of damaged soft coral is calculated as follows. The reference price of soft coral is 500 yuan / Kg. The protection coefficient of soft coral species is 1, and the involvement coefficient is 1.0. Therefore, the value of damaged soft coral in the damaged area is calculated as follows:

[0125] Table 2 Calculation of the value of damaged soft coral in the damaged area

[0126]

[0127] The value of damaged soft coral = the weight of damaged soft coral * the reference value of coral * the protection coefficient * the involvement coefficient

[0128] = 172.98 kg * 500 yuan / kg * 1 * 1 = 86490.00 yuan

[0129] The value of damaged soft coral is calculated to be about 86490.00 yuan.

[0130] Calculation of the value of damaged coral reef

[0131] The reference price of reef, as a derivative of coral, is 500 yuan / Kg. After being hit, the type of reef changes from original reef to coral reef, and the value coefficient of damaged coral reef decreases from 0.5 to 0.4, with a loss of 0.1. Therefore, the value of damaged reef in the damaged area is calculated as follows:

[0132] Table 3 Calculation of the value of damaged reef in the damaged area

[0133]

[0134]

[0135] The value of damaged coral reef = the weight of damaged coral reef * the reference value of coral * the involvement coefficient (the value coefficient of live reef - the value coefficient of reef fragments)

[0136] = 63040.28 kg * 500 yuan / kg * 0.1 = 3152014.00 yuan

[0137] The value of damaged coral reef resources is calculated to be about 3152014.00 yuan.

[0138] The total value of damaged coral reef resources is the sum of the value of damaged reef-building coral, the value of damaged soft coral, and the value of damaged coral reef stones. Therefore, the value of damaged coral reef resources in the area where the YYY ran aground is: value of damaged coral reef resources = value of damaged reef-building coral + value of damaged soft coral + value of damaged coral reef stones = 324,400.00 yuan + 86,490.00 yuan + 3,152,014.00 yuan = 3,562,904.00 yuan

[0139] The calculated value of damaged coral reef resources is approximately 3,562,904.00 yuan. The compensation for the loss of ecological services is calculated based on data collected and reference to the main report of The Economics of Ecosystems and Biodiversity (KUMAR, 2010) (TEEB) and related supplementary materials. The value of coral reef ecosystem services is approximately 13-120 million US dollars per hectare per year, which is equivalent to 93.5-862.1 yuan per square meter per year. Considering the low health level of the coral reef ecosystem in the area involved, the ecological service value is conservatively calculated at 93.5 yuan per square meter per year.

[0140] The damaged coral reef stones on site are mainly developed from the iterative accumulation of Favites sp. The linear extension rate of Favites sp. is approximately 0.5-1.5 cm per year (Li et al., 2018). The average thickness of the damaged coral reef stones is 21.6 cm, which would take approximately 14.4-43.2 years to recover at the above-mentioned deposition rate. Therefore, the compensation for the loss of ecological services is equal to the product of the damaged area, the annual service value of coral reef ecosystem per unit area, and the recovery period. Specifically:

[0141] Compensation for the loss of ecological services = damaged coral reef resource area * annual service value of coral reef ecosystem per unit area * recovery period

[0142] = 126.5 m2*93.5 yuan / m 2 / yr*(14.4-43.2) yr = 170,319.60-510,958.80 yuan

[0143] The calculated compensation for the loss of ecological services of this event is approximately 170,319.60-510,958.80 yuan.

[0144] According to the provisions of DB46 / T 584-2013 "Hainan Province Marine Ecological Damage Compensation and Ecological Compensation Evaluation Method", the repair cost of natural marine biological resources should be considered when calculating the loss value, and in principle, it should not be less than 3 times the direct economic loss. In this accident, both the reef-building coral and the soft coral belong to natural marine organisms, so the resource damage repair cost of the two should not be less than 1232670.00 yuan. Coral reef is not a natural marine organism, but it is not only the base stone of coral growth, but also an important part of land resources. Its damage repair cost should not be less than 1 times the damaged value, which is 3152014.00 yuan. Therefore, the resource damage repair cost of this accident should be 4384684.00 yuan.

[0145] Obviously, the above embodiments are only examples for clearly illustrating but not limiting the implementation. Based on the above description, other different forms of changes or variations can be made by those skilled in the art. Here, all the implementations are not required to be exhausted, and the obvious changes or variations derived therefrom are still within the protection scope of the present application.

Claims

1. An intelligent assistant system for coral reef impairment assessment, characterized by, The method comprises the following steps: An imaging acquisition module is used to acquire images of the coral reef through an underwater camera; An identification module comprises a trained convolutional neural network, which is trained based on the species of the coral and the live or dead state of the coral, and is used to input the images of the coral reef and output the proportion, coverage, area and mortality rate of each species of the coral reef; A density detection module is used to measure the density of the acquired coral reef sample, and the measurement is based on the identification of the species of the coral reef by the identification module, and the average density of multiple samples of a fixed species of the coral reef is calculated; An estimation module is used to input the field evaluation of the damaged thickness, and calculate the weight of the damaged coral reef based on the area and density of the coral reef, and the value of each species of the coral is calculated based on the formula: VAL = M * P * K1 * K2, wherein VAL is the value of the damaged coral reef, M is the weight of the damaged coral reef, P is the reference price of the coral species, K1 is the protection value coefficient of the coral, and K2 is the rare value coefficient of the coral; The values of each species of the coral are added to obtain the economic loss value for evaluating the damage of the coral reef.

2. The intelligent assistant system for coral reef impairment assessment of claim 1, wherein, The identification module is also used to distinguish soft corals, reef-building stony corals and reefs, and the density detection module and the estimation module are used to independently detect the density and estimate the value of the soft corals, the reef-building stony corals and the reefs.

3. The intelligent assistant system for coral reef impairment assessment of claim 2, wherein, The density detection module acquires the weight of the coral reef sample based on weighing, and calculates the volume of the irregular coral reef sample by the drainage method.

4. The intelligent assistant system for coral reef impairment assessment of claim 3, wherein, In the density detection module: The density of the reef-building stony coral = weight / drained volume, and the volume = height * area * effective proportion coefficient, wherein the effective proportion coefficient = drained volume / maximum length-width-height volume of the coral colony, in order to quickly calculate, the effective proportion coefficient of the massive coral colony is 0.8, the effective proportion coefficient of the skin shell-shaped coral colony is 0.6, the effective proportion coefficient of the leaf-shaped coral colony is 0.4, and the effective proportion coefficient of the branch-shaped coral colony is 0.2; The density of the soft coral = weight / drained volume, and the volume = height * area * effective proportion coefficient, wherein the effective proportion coefficient = drained volume / maximum length-width-height volume of the coral colony, in order to quickly calculate, the effective proportion coefficient of the table cover-shaped soft coral colony is 0.8, the effective proportion coefficient of the disc-shaped soft coral colony is 0.6, the effective proportion coefficient of the flower-shaped soft coral colony is 0.4, the effective proportion coefficient of the branch-shaped soft coral colony is 0.2, and the effective proportion coefficient of the fan-shaped soft coral colony is 0.

1.

5. The intelligent assistant system for coral reef impairment assessment of claim 4, wherein, The value formula of the reef is reef weight * reef reference value * value damage coefficient, wherein the value damage coefficient is the reef value coefficient - reef fragment value coefficient; The convolutional neural network is also trained based on the damage reason of the coral reef, the proportion of each species of the coral reef, the mortality rate of the damaged coral reef, the shape of the coral reef and the evaluated damaged thickness, and the convolutional neural network is used to input the damage reason of the coral reef and output the damaged thickness based on the images of the coral reef; and the imaging acquisition module is also used to denoise the images of the coral reef.

6. The intelligent assistant system for coral reef impairment assessment of claim 5, wherein, The estimation module is also used to calculate the value of the damaged reef-building stony coral, the value of the damaged soft coral and the value of the damaged coral reef, and add the above three values to obtain the total economic loss value of the damaged coral reef.

7. The intelligent assistant system for coral reef impairment assessment of claim 6, wherein, The estimation module is also used to calculate the ecological service function loss compensation fee, which is calculated based on the area of the damaged coral reef, the annual service value of the coral reef ecosystem per unit area and the recovery period of the damaged coral reef.

8. The intelligent assistant system for coral reef impairment assessment of claim 7, wherein, The estimation module is further configured to calculate a resource damage repair cost, the resource damage repair cost including a repair cost of damaged reef-building coral, a repair cost of damaged soft coral, and a repair cost of damaged coral reef; wherein the repair costs of the damaged reef-building coral and the damaged soft coral are calculated based on their respective damaged values by a certain multiple, and the repair cost of the damaged coral reef is determined based on its damaged value and in combination with expert recommendations.

9. An intelligent assistance method for coral reef damage assessment, based on the method of the intelligent assistance system for coral reef damage assessment according to claim 8, characterized in that, Comprise: Step one, collect coral reef images for training, the coral reef images including collecting undamaged coral reef images and damaged coral reef images in each sea area by using underwater cameras, and obtaining the damage causes, labeling the coral reef images and training a convolutional neural network; Step two, determine the affected range according to the damage event, ride a waterborne vehicle to the damage event, and release an underwater camera to collect underwater coral reef images in the affected range, and salvage or collect coral reef samples; Step three, input the coral reef images into the convolutional neural network, and identify and detect the density of the coral reef samples, combine the density information and the convolutional neural network output result, calculate the lost soft coral, reef-building coral and reef values respectively, and add the soft coral, reef-building coral and reef values to obtain an economic loss value for evaluating the coral reef damage.

10. The intelligent assistance method for coral reef impairment assessment according to claim 9, wherein, In step two, when the underwater camera collects images, a 50m-100m scale tape is laid on the collected section, and an underwater digital camera is used to shoot along the scale tape from one end of the section.

Images