Marine culture area extraction method based on SAR (Synthetic Aperture Radar) image

Abstract

The invention provides a marine culture area extraction method based on an SAR image, and the method comprises the following steps: S1, screening out a preprocessed SAR image of a target area in an image library, extracting features from the SAR image, and stacking the features into a feature image; s2, performing initial segmentation on the feature image by using a watershed algorithm; s3, on the basis of the initial segmentation in the S2, performing region merging by using a similarity criterion to obtain a final segmentation result; s4, in combination with the SAR image, marking the type of the offshore culture area on the final segmentation result; s5, dividing the samples of the offshore culture area into a training set and a verification set according to a preset proportion; s6, evaluating the trained DeepLabV3 model by using a verification set; and S7, carrying out offshore culture area extraction on the SAR image to be detected by using the verified DeepLabV3 model. The marking burden is relieved in a semi-automatic mode, and automatic monitoring and classification of the offshore culture area are achieved step by step.

Description

Technical Field

[0001] The present invention relates to the field of smart maritime technology, and in particular to a method for extracting offshore aquaculture areas based on SAR images. Background Art

[0002] Marine aquaculture areas include cage aquaculture, raft aquaculture, cylindrical structure aquaculture and other types, and their production scale and number are increasing year by year. These aquaculture areas play an important role in the marine ecosystem, and also have a profound impact on the marine environment and shipping safety. Therefore, the precise extraction and classification of marine aquaculture areas has important scientific research and practical application value.

[0003] Traditional methods for monitoring offshore aquaculture areas mainly rely on field surveys and high-resolution optical remote sensing images. However, field surveys have disadvantages such as high cost, low efficiency, and limited spatial coverage. Although optical remote sensing images can provide high-resolution spatial information, they are easily affected by weather conditions, light changes, and cloud cover, making it difficult to achieve all-day and all-weather monitoring. Synthetic Aperture Radar (SAR), as an active microwave remote sensing technology, has all-day and all-weather monitoring capabilities. SAR images can provide rich information such as the intensity, texture, shape, and polarization of the target, giving it significant advantages in marine target detection and recognition. However, due to the complex imaging mechanism of SAR images, high noise, and the complex and diverse structure of offshore aquaculture areas, the extraction and classification of offshore aquaculture areas based on SAR images has become a challenging task.

[0004] In recent years, deep learning technology has made breakthrough progress in the field of computer vision. Driven by big data, it can automatically learn the deep features of images, greatly improving the performance of target detection and recognition. Target detection methods based on convolutional neural networks and other deep learning models have gradually been applied to SAR image processing. However, deep learning methods rely on a large number of labeled samples, and the labeling process of SAR images is often time-consuming and laborious, especially in the extraction and classification of complex targets such as offshore aquaculture areas. Summary of the invention

[0005] In view of this, the purpose of the present invention is to propose a method for extracting offshore aquaculture areas based on SAR images, which aims to reduce the labeling burden in a semi-automatic way and gradually realize the automated monitoring and classification of offshore aquaculture areas.

[0006] The technical means adopted by the present invention are as follows:

[0007] A method for extracting offshore aquaculture areas based on SAR images comprises the following steps:

[0008] S1. In the image library, select the pre-processed SAR images of the target area, extract features from the SAR images, and stack the features into feature images;

[0009] S2, use the watershed algorithm to perform initial segmentation on the feature image;

[0010] S3, based on the initial segmentation in S2, the similarity criterion is used to merge the regions to obtain the final segmentation result;

[0011] S4. Combined with the SAR image, the types of offshore aquaculture areas are marked on the final segmentation results. The types of offshore aquaculture areas include cage aquaculture areas, raft aquaculture areas and cylindrical structure aquaculture areas, and the corresponding masks are obtained;

[0012] S5. Using the labeled mask and the feature image to train a Bayesian classifier, using the trained Bayesian classifier to classify the offshore aquaculture areas, post-processing the classified areas to obtain offshore aquaculture area samples, and dividing the offshore aquaculture area samples into a training set and a validation set according to a preset ratio;

[0013] S6, input the training set into the DeepLabV3 model for training, use the validation set to evaluate the trained DeepLabV3 model, if the evaluation accuracy does not meet the standard, return to S4 for step-by-step iteration; if the evaluation accuracy meets the standard, output the verified DeepLabV3 model and execute S7;

[0014] Preferably, the standard may use accuracy and recall, and when both accuracy and recall are greater than 85%, S7 is executed;

[0015] S7. Use the verified DeepLabV3 model to extract offshore aquaculture areas from the SAR images to be detected.

[0016] Furthermore, in S1, extracting features from SAR images includes the following steps:

[0017] S11. The texture features of the SAR image are calculated through the gray level co-occurrence matrix, including energy, contrast and entropy. The calculation formula is as follows:

[0018]

[0019]

[0020] Where P(i,j) represents the value of position (i,j) in the gray level co-occurrence matrix, and N is the number of gray levels;

[0021] S12, polarization characteristics include normalized polarization index NPDI, scattering entropy H and average scattering angle α, and the calculation formula is as follows:

[0022]

[0023] Among them, p i is the normalization of the eigenvalues ​​of the coherence matrix;

[0024]

[0025] Among them, α i is the scattering angle corresponding to the eigenvalue of the coherence matrix.

[0026] Furthermore, in S3, the specific steps of using the similarity criterion to merge regions are as follows:

[0027] S31, by calculating the similarity of the texture features of two adjacent regions, using Euclidean distance as a similarity measurement method, assuming that the texture feature vectors of the two regions A and B are T A and T B , the similarity measure is:

[0028]

[0029] S32. For dual-polarization SAR images, calculate the similarity of polarization features of two regions, and set the polarization feature vector as P A and P B , the similarity measure is:

[0030]

[0031] S33, the overall evaluation function of merging two adjacent regions A and B is expressed as:

[0032] R(A,B)=w 1 ·d texture (A,B)+w 2 ·d polarization (A,B)

[0033] Among them, w 1 , w 2 is the weight coefficient, which is used to balance the influence of texture and polarization feature similarity;

[0034] S34. Using the preset parameters as the stopping condition, after the stopping condition is reached, the final segmentation result is obtained.

[0035] Furthermore, in S5, the steps of using the trained Bayesian classifier to classify the offshore aquaculture areas are as follows:

[0036] S51. Using the labeled mask and the feature image, a Bayesian classifier is trained. The trained Bayesian classifier classifies the offshore aquaculture areas. The classification process is as follows:

[0037]

[0038] in, is the predicted category, In order to make the posterior probability P(C k |x) Maximum category C k , P(C k ) is Category C k The prior probability, P(x|C k ) is in category C k Under the condition of , the likelihood probability of feature x appearing;

[0039] S52, rasterizing and vectorizing the results of the classification in S51, checking the types of offshore aquaculture areas and the boundaries of the patches, and obtaining offshore aquaculture area samples;

[0040] S53. Divide the offshore aquaculture area samples into a training set and a validation set according to a preset ratio.

[0041] Furthermore, the steps of stepwise iteration in S6 are as follows:

[0042] S61. In the initial stage, use the training set in S53, denoted as D 0 , train and obtain the initial DeepLabV3 model, denoted as M 0 ,but:

[0043]

[0044] Among them, θ is the parameter of the model, L is the loss function, and the calculation formula of L is as follows:

[0045]

[0046] in, is the network’s predicted probability for category c, p i (c) is the indicator function of the true category c;

[0047] S62, using the initial model M 0 Predict the data of the validation set in S53, and identify false positives and missed positives by analyzing the prediction results. Error analysis is calculated by calculating the confusion matrix:

[0048]

[0049] Among them, TP is the number of true positive cases, FP is the number of false positive cases, FN is the number of false negative cases, and TN is the number of true negative cases;

[0050] S63. According to the error analysis results, select those areas that are misclassified by the model, add samples, and generate a new data set D 1 , and retrain the model to obtain the improved model M1 ;

[0051] S64, repeat S62, S63, and continuously update the model. In each iteration k, there are:

[0052]

[0053] S65, setting threshold L val As the iteration termination condition, stop the iteration:

[0054] |L val,k -L val,k-1 |<ε

[0055] Among them, L val,k is the loss on the validation set at the kth iteration, and ε represents the change threshold.

[0056] The present invention also provides a storage medium, which includes a stored program, wherein when the program is run, any of the above-mentioned methods for extracting offshore aquaculture areas based on SAR images is executed.

[0057] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes any one of the above-mentioned methods for extracting marine aquaculture areas based on SAR images through the operation of the computer program.

[0058] Compared with the prior art, the present invention has the following advantages:

[0059] Fast modeling: By using the Bayesian classifier for preliminary classification, the present invention can quickly build a model in the initial stage with only a small number of labeled samples. The classifier performs well in small samples and high-dimensional feature spaces, can effectively use limited labeled data, quickly obtain preliminary classification results, reduce the initial demand for large-scale data, and reduce the burden of labeling.

[0060] The present invention first uses traditional machine learning methods for preliminary classification, and then uses deep learning models for refined training to continuously improve the model. Each round of iteration can improve the performance of the model and gradually improve the detection and classification capabilities of aquaculture areas. By gradually expanding the training set and adopting technologies such as data enhancement, the generalization ability of the deep learning model is improved, and it can better adapt to offshore aquaculture areas of different types and scales. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0062] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0063] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0064] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0065] like Figure 1 As shown, the present invention provides a method for extracting offshore aquaculture areas based on SAR images, comprising the following steps:

[0066] S1. In the image library, select the preprocessed SAR images of the target area, extract features from the images, and stack the features into feature images; the preprocessing includes complex data conversion, EnLee filtering, geocoding and other steps.

[0067] In the remote sensing image database, 200 pre-processed SAR images of the GF-3 satellite FSI imaging mode in the eastern waters of Liaoning Province and the eastern waters of Fujian Province from January to December were selected.

[0068] The feature extraction process is as follows:

[0069] S11 calculates the texture features of SAR images through the gray-level co-occurrence matrix, including energy, contrast, and entropy. The calculation formula is as follows:

[0070]

[0071] Where P(i,j) represents the value of position (i,j) in the gray level co-occurrence matrix, and N is the number of gray levels;

[0072] S12, polarization characteristics include normalized polarization index (NPDI), scattering entropy (H) and average scattering angle (α), and the calculation formula is as follows:

[0073]

[0074] Among them, p i is the normalization of the eigenvalues ​​of the coherence matrix;

[0075]

[0076] Among them, α i is the scattering angle corresponding to the eigenvalue of the coherence matrix.

[0077] S2, use the watershed algorithm to perform initial segmentation on the feature image;

[0078] S3, based on the initial segmentation, the similarity criterion is used to merge the regions to obtain the final segmentation result;

[0079] The process of region merging using similarity criteria is as follows:

[0080] S31, by calculating the similarity of the texture features of two adjacent regions, using Euclidean distance as a similarity measurement method, assuming that the texture feature vectors of the two regions A and B are T A and T B , the similarity measure is:

[0081]

[0082] S32. For dual-polarization SAR images, calculate the similarity of polarization features of two regions, and set the polarization feature vector as P A and P B , the similarity measure is:

[0083]

[0084] S33, the overall evaluation function for merging two adjacent regions A and B can be expressed as:

[0085] R(A,B)=w 1 ·dtexture (A,B)+w 2 ·d polarization (A,B)

[0086] S34 uses the preset parameters as the stopping condition, and obtains the final segmentation result after the stopping condition is met.

[0087] Among them, w 1 , w 2 is the weight coefficient used to balance the influence of texture and polarization feature similarity, w 1 Set to 0.4, w 2 Set to 0.6.

[0088] S4. Combined with the SAR image, the types of offshore aquaculture areas are marked on the final segmentation results, including cage aquaculture areas, raft aquaculture areas and cylindrical structure aquaculture areas, and the corresponding masks are obtained;

[0089] S5. Using the labeled mask and the feature image to train a Bayesian classifier, the trained Bayesian classifier classifies the offshore aquaculture areas, performs post-processing on the offshore aquaculture area samples, and divides the samples into a training set and a validation set according to a preset ratio;

[0090] The process of obtaining offshore aquaculture area samples by training the Bayesian classifier is as follows:

[0091] S51. Using the labeled mask and the feature image, a Bayesian classifier is trained. The trained Bayesian classifier classifies the offshore aquaculture areas. The classification process is as follows:

[0092]

[0093] in, is the predicted category, In order to make the posterior probability P(C k |x) Maximum category C k , P(C k ) is Category C k The prior probability, P(x|C k ) is in category C k Under the condition of , the likelihood probability of feature x appearing;

[0094] S52, rasterizing the above results, checking the types of offshore aquaculture areas and the boundaries of the spots, and obtaining offshore aquaculture area samples;

[0095] S53 is divided into training set and validation set in a ratio of 8:2;

[0096] S6, input the training set into the DeepLabV3 model for training, use the validation set to evaluate the model, if the accuracy is insufficient, return to S4, perform stepwise iterations until the accuracy meets the requirements and execute step S7;

[0097] The step-by-step iterative process is:

[0098] S61. In the initial stage, use the training set in S53, denoted as D 0 , train and obtain the initial DeepLabV3 model, denoted as M 0 ,but:

[0099]

[0100] Among them, θ is the parameter of the model, L is the loss function, and the calculation formula of L is as follows:

[0101]

[0102] in, is the network’s predicted probability for category c, p i (c) is the indicator function of the true category c;

[0103] S62, using the initial model M 0 Predict the data of the validation set in S53 and identify false positives and missed positives by analyzing the prediction results. Error analysis can be calculated by calculating the confusion matrix:

[0104]

[0105] Among them, TP is the number of true positive cases, FP is the number of false positive cases, FN is the number of false negative cases, and TN is the number of true negative cases;

[0106] S63. According to the error analysis results, select those areas that are misclassified by the model, add samples, and generate a new data set D 1 , and retrain the model to obtain the improved model M 1 ;

[0107] S64, repeat S62, S63, and continuously update the model. In each iteration k, there are:

[0108]

[0109] S65, setting threshold L val As the iteration termination condition, stop the iteration:

[0110] |L val,k -L val,k-1 |<ε

[0111] Among them, L val,kis the loss of the kth iteration on the validation set, and ε is a small positive number indicating the change threshold. ε is set to 0.001.

[0112] S7. Use the verified model to extract the offshore aquaculture area from the SAR image to be detected.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for extracting offshore aquaculture areas based on SAR images, characterized in that: The steps include: S1. In the image library, select the pre-processed SAR images of the target area, extract features from the SAR images, and stack the features into feature images; S2, use the watershed algorithm to perform initial segmentation on the feature image; S3, based on the initial segmentation in S2, the similarity criterion is used to merge the regions to obtain the final segmentation result; S4. Combined with the SAR image, the types of offshore aquaculture areas are marked on the final segmentation results. The types of offshore aquaculture areas include cage aquaculture areas, raft aquaculture areas and cylindrical structure aquaculture areas, and the corresponding masks are obtained; S5. Using the labeled mask and the feature image to train a Bayesian classifier, using the trained Bayesian classifier to classify the offshore aquaculture areas, obtaining offshore aquaculture area extraction results and performing post-processing to obtain offshore aquaculture area samples, and dividing the offshore aquaculture area samples into a training set and a validation set according to a preset ratio; S6, input the training set into the DeepLabV3 model for training, use the validation set to evaluate the trained DeepLabV3 model, if the evaluation accuracy does not meet the standard, return to S4 for step-by-step iteration; if the evaluation accuracy meets the standard, output the verified DeepLabV3 model and execute S7; S7. Use the verified DeepLabV3 model to extract offshore aquaculture areas from the SAR images to be detected.

2. The method for extracting marine aquaculture areas based on SAR images according to claim 1, characterized in that: In S1, extracting features from SAR images includes the following steps: S11. The texture features of the SAR image are calculated through the gray level co-occurrence matrix, including energy, contrast and entropy. The calculation formula is as follows: Where P(i,j) represents the value of position (i,j) in the gray level co-occurrence matrix, and N is the number of gray levels; S12, polarization characteristics include normalized polarization index NPDI, scattering entropy H and average scattering angle α, and the calculation formula is as follows: Among them, p i is the normalization of the eigenvalues ​​of the coherence matrix; Among them, α i is the scattering angle corresponding to the eigenvalue of the coherence matrix.

3. The offshore aquaculture area extraction method based on SAR images according to claim 1 is characterized in that: In S3, the specific steps of using the similarity criterion to merge regions are as follows: S31, by calculating the similarity of the texture features of two adjacent regions, using Euclidean distance as a similarity measurement method, assuming that the texture feature vectors of the two regions A and B are T A and T B , the similarity measure is: S32. For dual-polarization SAR images, calculate the similarity of polarization features of two regions, and set the polarization feature vector as P A and P B , the similarity measure is: S33, the overall evaluation function of merging two adjacent regions A and B is expressed as: R(A,B)=w1·d texture (A,B)+w2·d polarization (A,B) Among them, w1 and w2 are weight coefficients used to balance the influence of texture and polarization feature similarity; S34. Using the preset parameters as the stopping condition, after the stopping condition is reached, the final segmentation result is obtained.

4. The method for extracting marine aquaculture areas based on SAR images according to claim 1, characterized in that: In S5, the steps of using the trained Bayesian classifier to classify offshore aquaculture areas are as follows: S51. Using the labeled mask and the feature image, a Bayesian classifier is trained. The trained Bayesian classifier classifies the offshore aquaculture areas. The classification process is as follows: in, is the predicted category, In order to make the posterior probability P(C k |x) Maximum category C k , P(C k ) is Category C k The prior probability, P(x|C k ) is in category C k Under the condition of , the likelihood probability of feature x appearing; S52, rasterizing and vectorizing the results of the classification in S51, checking the types of offshore aquaculture areas and the boundaries of the patches, and obtaining offshore aquaculture area samples; S53. Divide the offshore aquaculture area samples into a training set and a validation set according to a preset ratio.

5. The method for extracting marine aquaculture areas based on SAR images according to claim 1, characterized in that: The steps of stepwise iteration in S6 are as follows: S61. In the initial stage, the training set in S53 is used, denoted as D0, and the initial DeepLabV3 model is obtained by training, denoted as M0, then: Among them, θ is the parameter of the model, L is the loss function, and the calculation formula of L is as follows: in, is the network’s predicted probability for category c, p i (c) is the indicator function of the true category c; S62, use the initial model M0 to predict the data of the validation set in S53, and identify false positives and missed positives by analyzing the prediction results. The error analysis is calculated by calculating the confusion matrix: Among them, TP is the number of true positive cases, FP is the number of false positive cases, FN is the number of false negative cases, and TN is the number of true negative cases; S63. According to the error analysis results, select those areas that are misclassified by the model, add samples, generate a new data set D1, and retrain the model to obtain an improved model M1; S64, repeat S62, S63, and continuously update the model. In each iteration k, there are: S65. Setting threshold L val As the iteration termination condition, stop the iteration: |L val,k -L val,k-1 |<e Among them, L val,k is the loss on the validation set at the kth iteration, and ε represents the change threshold.

6. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is run, the offshore aquaculture area extraction method based on SAR images as described in any one of claims 1 to 5 is executed.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes the offshore aquaculture area extraction method based on SAR images as described in any one of claims 1 to 5 through the operation of the computer program.

Images