Deep learning-based duckweed type water body remote sensing monitoring method and device

Abstract

The invention provides a duckweed type water body remote sensing monitoring method and device based on deep learning, and the method comprises the steps: obtaining a remote sensing tile of a duckweed type water body, and taking the remote sensing tile as an RGB three-channel sample; calculating the RGB three-channel sample based on a chromaticity angle, a slope sum and a green channel index to obtain a chromaticity angle wave band, a slope sum wave band and a green channel index wave band, and performing wave band synthesis to obtain an ASGI remote sensing image tile; setting an attribute threshold according to the duckweed type water body in the ASGI remote sensing image tile, and constructing a sample data set of the duckweed type water body; and training an initial deep learning model through the sample data set to obtain a duckweed type water body recognition model so as to recognize the duckweed type water body. Through the method and the device, the technical problem that the recognition accuracy is reduced due to the defects that part of duckweed-type water bodies are small in area and easy to confuse with other ground features when the duckweed-type water bodies are recognized by using a remote sensing means in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of water environment remote sensing technology, and in particular to a duckweed-type water body remote sensing monitoring method and device based on deep learning. Background Art

[0002] In addition to the typical black and smelly water bodies in cities, new forms of duckweed and water blooms have appeared in rural black and smelly water bodies. Among them, duckweed plants are the smallest monocotyledonous flowering plants in the world, with fast growth and strong tolerance. There are mainly 4 genera and about 10 species. A large number of studies have shown that many types of duckweed have the ability to remove nitrogen and phosphorus nutrients from water bodies and are widely used in water purification processes. However, when a large number of duckweeds gather or burst, covering the entire or part of the water surface, it will cause hypoxia and eutrophication of the water body, hindering the reoxygenation of the water body and the aquatic organisms from receiving light, and the aquatic organisms will die due to lack of oxygen, and the water quality of the water body will gradually decline. There are few studies on the reaction of duckweed to water quality after it becomes a dominant population. At the same time, the outbreak of harmful algae blooms has long been an ecological and environmental problem faced by freshwater lakes and ponds around the world.

[0003] There are three possibilities for water bodies covered by duckweed. First, the water quality is normal, duckweed floats on the surface, and has not formed a population advantage. Duckweed can be salvaged for livestock and poultry food; second, the water quality is poor, and duckweed is artificially raised to purify the water quality of the water body; third, black and smelly water bodies, duckweed covers all or most of the water surface, forming a population advantage, hindering the growth of other photosynthetic plants and animals in the water, causing the death of related organisms, and the water body becomes black and smelly. Therefore, it is very important to effectively identify duckweed-type water bodies. However, since duckweed-type water bodies are mostly distributed in rural areas, with a large distribution area and scattered distribution, it is difficult to identify duckweed-type water bodies through ground investigation methods. The identification of duckweed-type water bodies by remote sensing means is mainly to distinguish the physical characteristics of duckweed-type water bodies from other landforms. Although it has the characteristics of fast and large range, it also requires higher resolution of remote sensing images, and some duckweed-type water bodies are small in area and easily confused with other landforms, resulting in reduced recognition accuracy. Summary of the invention

[0004] The present invention provides a remote sensing monitoring method and device for duckweed-type water bodies based on deep learning, so as to solve the technical problem that when using remote sensing means to identify duckweed-type water bodies, the recognition accuracy is reduced due to the small area of ​​some duckweed-type water bodies, which are easily confused with other ground objects.

[0005] The present invention provides a method for remote sensing monitoring of duckweed-type water bodies based on deep learning, comprising the following steps: Get the remote sensing tiles of duckweed-type water bodies as RGB three-channel samples; Performing calculation based on the chromaticity angle on the RGB three-channel samples to obtain a chromaticity angle band, performing calculation based on the slope sum on the RGB three-channel samples to obtain a slope sum band, and performing calculation based on the green channel index on the RGB three-channel samples to obtain a green channel index band; Performing band synthesis on the chromaticity angle band, the slope sum band and the green channel index band to obtain an ASGI remote sensing image tile; Setting attribute thresholds according to the duckweed-type water bodies in the ASGI remote sensing image tiles, and constructing a sample data set of duckweed-type water bodies; The initial deep learning model is trained by the sample data set to obtain a duckweed-type water body recognition model to identify duckweed-type water bodies.

[0006] In some embodiments, the step of performing calculation based on the chromaticity angle on the RGB three-channel samples to obtain the chromaticity angle band includes: Get the band linear summation coefficient of map remote sensing; According to the band linear summation coefficient, the R value, the G value and the B value of the RGB three-channel sample are subjected to stimulus value calculation to obtain the XYZ three primary color stimulus values; Based on the XYZ three primary color stimulus values, determine the chromaticity abscissa and the chromaticity ordinate; The chromaticity angle is calculated for the chromaticity horizontal coordinate and the chromaticity vertical coordinate to obtain a chromaticity angle band.

[0007] In some embodiments, the step of performing the calculation based on the slope and the band on the RGB three-channel samples to obtain the slope and the band includes: Determine a first ratio of the B value to the G value in the RGB three-channel sample; Determine a second ratio of the R value to the G value in the RGB three-channel sample; The sum of the first ratio and the second ratio is taken as the slope and band.

[0008] In some embodiments, the green channel index band is obtained by calculating the green channel index based on the RGB three-channel samples, including: Determine the sum of the channel values ​​of the B value and the R value in the RGB three-channel sample; The ratio of the square of the G value in the RGB three-channel sample to the sum of the channel values ​​is used as the green channel index band.

[0009] In some embodiments, setting an attribute threshold according to the duckweed-type water body in the ASGI remote sensing image tile and constructing a sample data set of the duckweed-type water body includes: Determine the surface vector of the duckweed-type water body in the ASGI remote sensing image tile; When the duckweed-type water body is a real duckweed-type water body, setting a first attribute threshold for the surface vector of the duckweed-type water body; When the duckweed type water body is a non-duckweed type water body, setting a second attribute threshold for the surface vector of the duckweed type water body; The data label corresponding to the ASGI remote sensing image tile is constructed according to the first attribute threshold and the second attribute threshold, and a sample data set of duckweed-type water bodies is obtained.

[0010] In some embodiments, the initial deep learning model is trained by the sample data set to obtain a duckweed-type water body recognition model, including: Inputting the training samples of the sample data set into the initial deep learning model for forward propagation to obtain the predicted attribute values ​​of the training samples for duckweed-type water bodies; Constructing a training loss value according to the difference between the predicted attribute value and the data label of the training sample, and back-propagating the training loss value in the initial deep learning model to update the model parameters of the initial deep learning model; When the training loss value begins to converge or reaches a preset training round, the training of the initial deep learning model is terminated to obtain a duckweed-type water body recognition model.

[0011] The present invention also provides a duckweed-type water body remote sensing monitoring device based on deep learning, which includes the following modules: The acquisition module is used to obtain the remote sensing tiles of duckweed-type water bodies as RGB three-channel samples; A determination module, configured to perform a calculation based on a chromaticity angle on the RGB three-channel samples to obtain a chromaticity angle band, perform a calculation based on a slope sum on the RGB three-channel samples to obtain a slope sum band, and perform a calculation based on a green channel index on the RGB three-channel samples to obtain a green channel index band; A synthesis module, used for synthesizing the chromaticity angle band, the slope sum band and the green channel index band to obtain an ASGI remote sensing image tile; A construction module, used to set attribute thresholds according to the duckweed-type water bodies in the ASGI remote sensing image tiles, and to construct a training data set for the duckweed-type water bodies; The recognition module is used to train the initial deep learning model through the training data set to obtain a duckweed-type water body recognition model to identify duckweed-type water bodies.

[0012] 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 when the processor executes the computer program, the remote sensing monitoring method for duckweed-type water bodies based on deep learning as described in any one of the above-mentioned methods is implemented.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the remote sensing monitoring method for duckweed-type water bodies based on deep learning as described in any one of the above is implemented.

[0014] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the deep learning-based remote sensing monitoring methods for duckweed-type water bodies as described above.

[0015] The present invention provides a method and device for remote sensing monitoring of duckweed-type water bodies based on deep learning. Based on the remote sensing tiles of duckweed-type water bodies, three physical features, namely, chromaticity angle band, slope and band, and green channel index band, are constructed to identify duckweed-type water bodies. The three physical features are combined to obtain ASGI remote sensing image tiles, and a training data set of duckweed-type water bodies is constructed by setting attribute thresholds to train an initial deep learning model, thereby obtaining a duckweed-type water body recognition model to identify duckweed-type water bodies. Thus, duckweed-type water bodies are simultaneously identified by the three physical features of chromaticity angle band, slope and band, and green channel index band, which can eliminate the confusing effects of other objects in a targeted manner, improve the recognition accuracy, and overcome the defect that the resolution of remote sensing images in the prior art is not high, resulting in a reduced recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced one by one 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 work.

[0017] Figure 1 It is a flow chart of the duckweed-type water body remote sensing monitoring method based on deep learning provided by the present invention.

[0018] Figure 2 This is a framework schematic diagram of the deep learning-based duckweed-type water body remote sensing monitoring method provided by the present invention.

[0019] Figure 3 This is a schematic diagram of distinguishing easily confused objects after ASGI band combination provided by the present invention.

[0020] Figure 4 It is a structural schematic diagram of the duckweed-type water body remote sensing monitoring device based on deep learning provided by the present invention.

[0021] Figure 5 It is a schematic diagram of the physical structure of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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 are within the scope of protection of the present invention.

[0023] Combine the following Figure 1 Describe the duckweed-type water body remote sensing monitoring method based on deep learning of the present invention,

[0024] Figure 1 is a flow chart of the duckweed-type water body remote sensing monitoring method based on deep learning provided by the present invention, such as Figure 1 As shown, the method includes the following steps 101 to 105.

[0025] Step 101: obtain remote sensing tiles of duckweed-type water bodies as RGB three-channel samples.

[0026] First, we need to obtain remote sensing tiles of duckweed-type water bodies. The remote sensing tiles are specifically Google Earth remote sensing image tiles with high spatial resolution. After preprocessing, they are used as RGB (Red Green Blue) three-channel samples. During the preprocessing process, remote sensing images with different colors and many clouds are removed, and remote sensing images with uniform colors and few clouds are retained.

[0027] Step 102: perform calculation based on the chromaticity angle on the RGB three-channel samples to obtain the chromaticity angle band, perform calculation based on the slope sum on the RGB three-channel samples to obtain the slope sum band, and perform calculation based on the green channel index on the RGB three-channel samples to obtain the green channel index band.

[0028] See here Figure 2 , Figure 2 This is a schematic diagram of the framework of the method for remote sensing monitoring of duckweed-type water bodies based on deep learning provided by the present invention. After obtaining Google Earth remote sensing image tiles as RGB three-channel samples, in the ASGI band calculation and synthesis module, the embodiment of the present invention calculates the three physical characteristics of the chromatic angle band (a), slope sum band (S), and green channel index band (GI) of the RGB three-channel samples to distinguish duckweed-type water bodies. Among them, ASGI represents the abbreviation of the combination of three bands: chromatic angle (A), slope sum (S), and green channel index (GI). The calculation order of these three bands is not in particular order. The embodiment of the present invention has a distinction in order to explain them one by one, and there should be no such limitation in actual implementation.

[0029] For the chromaticity angle band a, the RGB three-channel samples are calculated based on the chromaticity angle to obtain the chromaticity angle band. First, the band linear summation coefficients of the map remote sensing are obtained. These coefficients are fixed parameters in the chromaticity angle algorithm. Then, according to the band linear summation coefficients, the R value, G value and B value of the RGB three-channel samples are calculated for the stimulus value to obtain the XYZ three primary color stimulus values, which are expressed as the following formulas (1) to (3): (1)

[0030] (2)

[0031] (3)

[0032] In the above formulas (1) to (3), R, G, and B represent the R value, G value, and B value of the RGB three-channel samples respectively, and the other fixed parameters are the band linear summation coefficients of map remote sensing.

[0033] Next, based on the XYZ three primary color stimulus values, the chromaticity coordinates are calculated, that is, the chromaticity horizontal coordinate x and the chromaticity vertical coordinate y are determined. The calculation formula is expressed as follows: Formula (4) to Formula (5): (4)

[0034] (5)

[0035] In the above formulas (4) and (5), X, Y, and Z represent the stimulus values ​​of the three primary colors, respectively.

[0036] Finally, the chromaticity angle is calculated for the chromaticity horizontal coordinate x and the chromaticity vertical coordinate y to obtain the chromaticity angle band a. The chromaticity angle calculation method is as follows: (6)

[0037] In the above formula (6), , They represent the calibration values ​​of the chromaticity horizontal coordinate x and the chromaticity vertical coordinate y respectively.

[0038] In the embodiment of the present invention, the chromaticity angle is used as a physical feature for identifying chromaticity angle duckweed-type water bodies and other landforms, considering that the chromaticity angle has a better effect on distinguishing duckweed-type water bodies from light green vegetation, red paint, bright blue paint and bright green paint. Because the chromaticity angle of duckweed-type water bodies is mainly distributed between 181.6°-182.0°, the chromaticity angles of light green vegetation (chromaticity angle of 181.1°-181.8°) and red paint (chromaticity angle of 179.7°-180.2°) are smaller than duckweed-type water bodies, and the chromaticity angles of bright blue paint (chromaticity angle of 181.9°-182.9°) and bright green paint (chromaticity angle of 182.7°-183.2°) are larger than duckweed-type water bodies. Therefore, the chromaticity angle can effectively distinguish easily confused landforms such as red paint and bright green paint.

[0039] As for the slope and band S, the RGB three-channel samples are calculated based on the slope sum to obtain the slope and band. The slope sum is calculated for the R value, G value, and B value in the RGB three-channel samples. First, the first ratio of the B value to the G value in the RGB three-channel samples is determined, and then the second ratio of the R value to the G value in the RGB three-channel samples is determined. Finally, the sum of the first ratio and the second ratio is taken as the slope and band S, which is expressed as the following formula (7): (7)

[0040] In the above formula (7), R, G, and B represent the R value, G value, and B value of the RGB three-channel sample, respectively.

[0041] In the embodiment of the present invention, the slope and the sum are used as physical features for identifying the chromatic angle duckweed-type water body and other landforms, considering that the slope and the sum can better distinguish the duckweed-type water body and the dark green paint. The slope and the sum of the duckweed-type water body are mainly in the range of 1.3-1.6, however, the slope and the sum of the dark green paint are mostly less than 1.3 or greater than 1.6, therefore, the slope and the sum can distinguish the duckweed-type water body and the dark green paint to a certain extent.

[0042] For the green channel index GI, the RGB three-channel samples are calculated based on the green channel index to obtain the green channel index band. Similarly, the calculation of the green channel index band is also based on the calculation of the R value, G value, and B value in the RGB three-channel samples. First, the sum of the channel values ​​of the B value and the R value in the RGB three-channel samples is determined, and then the ratio of the square of the G value in the RGB three-channel samples to the sum of the channel values ​​is taken as the green channel index band, which is expressed as the following formula (8): (8)

[0043] In the above formula (8), R, G, and B represent the R value, G value, and B value of the RGB three-channel sample, respectively.

[0044] In the embodiment of the present invention, the green channel index is used as a physical feature for identifying chromatic angle duckweed-type water bodies and other landforms, considering that the green channel index can effectively distinguish duckweed-type water bodies from sparse grasslands, green water bodies, and artificial grasslands. The green channel index of duckweed-type water bodies is mostly distributed between 105.1 and 136.9, the green channel index of sparse grasslands is less than 105, and the green channel index of green water bodies and artificial grasslands is mostly less than 105. Therefore, the green channel index has a good effect on distinguishing sparse grasslands, green water bodies, and artificial grasslands.

[0045] Step 103: synthesize the chromaticity angle band, the slope sum band, and the green channel index band to obtain an ASGI remote sensing image tile.

[0046] After the chromaticity angle band (a), the slope sum band (S), and the green channel index band (GI) of the RGB three-channel samples are calculated respectively through step 102, the chromaticity angle band, the slope sum band, and the green channel index band can be band synthesized to obtain an ASGI remote sensing image tile.

[0047] Step 104: set attribute thresholds according to the duckweed-type water bodies in the ASGI remote sensing image tiles, and construct a training data set for the duckweed-type water bodies.

[0048] like Figure 2 As shown, after the ASGI remote sensing image tiles are obtained through the ASGI band calculation and synthesis module, the next step is the model training stage. The embodiment of the present invention uses a recognition model based on deep learning to accurately identify duckweed-type water bodies. First, a duckweed-type water body recognition model needs to be trained to learn the physical characteristics of the above three bands. The initial deep learning model can use models such as Mobilenet-Uet, MobileNet, SqueezeNet, ShuffleNet, etc. The embodiment of the present invention uses the Mobilenet-Uet model as an example for illustration.

[0049] Here, we set attribute thresholds based on the duckweed-type water bodies in the ASGI remote sensing image tiles, and construct a sample data set of duckweed-type water bodies. First, we need to label the duckweed water bodies for the training data set, determine the surface vectors of the duckweed-type water bodies in the ASGI remote sensing image tiles, and then construct the duckweed water body sample set. In order to better train the initial deep learning model and make the recognition effect convincing, the number of sample data sets constructed is required to be no less than 1,000.

[0050] A planar shp file (also known as Shapefile file) is constructed for each ASGI remote sensing image tile, and then the duckweed-type water body is manually outlined in detail in this shp file, and each duckweed-type water body is outlined in the form of a planar vector. Subsequently, in the ArcGIS software, the surface vector is edited and the attribute value is set, that is, the duckweed-type water body and the background objects are distinguished by marking the attribute value. Specifically, when the duckweed-type water body is a real duckweed-type water body, a first attribute threshold is set for the planar vector of the duckweed-type water body, for example, 255. When the duckweed-type water body is a non-duckweed-type water body (such as a background object), a second attribute threshold is set for the planar vector of the duckweed-type water body, for example, 0.

[0051] Finally, the data labels corresponding to the ASGI remote sensing image tiles are constructed according to the first attribute threshold of 255 and the second attribute threshold of 0, and a sample data set of duckweed-type water bodies is obtained. Here, the marked first attribute threshold of 255 and the second attribute threshold of 0 can be directly used as the data labels corresponding to the ASGI remote sensing image tiles. After marking the attribute values ​​of each shp file, these shp files are converted into tif files, and the naming method corresponds to the name of the ASGI remote sensing image tile, so that the data labels corresponding to the ASGI remote sensing image tile can be matched according to the file name. Finally, all ASGI remote sensing image tiles and corresponding data labels are cut according to the same cutting rules, for example, they are uniformly cut into 512*512 pixel blocks as the sample data set of duckweed-type water bodies.

[0052] Step 105: Train the initial deep learning model using the sample data set to obtain a duckweed-type water body recognition model to identify duckweed-type water bodies.

[0053] like Figure 2 As shown, after the duckweed water body sample set is constructed, it is necessary to divide the duckweed water body sample set, that is, to divide the constructed sample data set. The embodiment of the present invention divides the sample data set into a training set, a validation set, and a test set according to a ratio of 8:1:1 or 7:2:1, wherein the training set is used to train the initial deep learning model Mobilenet-Uet, the validation set is used to verify the effect of the trained Mobilenet-Uet, and the test set is used to test and evaluate the recognition effect of Mobilenet-Uet. At the same time, it is necessary to ensure that the ratio of positive samples to negative samples in each type of divided data set is close to 1:1.

[0054] The following describes the process of training the initial deep learning model Mobilenet-Uet through the sample data set to obtain the duckweed-type water body recognition model. The training samples (training set) of the sample data set are input into the initial deep learning model for forward propagation to obtain the predicted attribute values ​​of the training samples for duckweed-type water bodies. During the forward propagation process, the initial deep learning model learns the three physical characteristics of the duckweed-type water body in the training sample, namely the chromatic angle band (a), slope and band (S), and green channel index band (GI), and predicts the predicted attribute values ​​of the duckweed-type water body based on the three physical characteristics. Then, according to the difference between the predicted attribute value and the data label of the training sample (that is, the labeled attribute value), the training loss value is constructed, and the training loss value is used to perform back propagation in the initial deep learning model. During the back propagation process, the gradient of the model is calculated and updated through the gradient optimization algorithm to update the model parameters of the initial deep learning model. When the training loss value begins to converge or reaches the preset training round, it means that the initial deep learning model has learned a set of optimal model parameters, and the training of the initial deep learning model ends. The initial deep learning model with the optimal model parameters can be used as the duckweed-type water body recognition model.

[0055] like Figure 2 As shown in the figure, after obtaining the optimal model parameters through training set training and then obtaining the duckweed type water body recognition model, the model prediction process of the accuracy verification stage is entered. Here, the duckweed water body sample set is divided into verification values ​​and test sets to verify the accuracy of the model (that is, the prediction performance). The duckweed type water body recognition model can be used to identify the duckweed type water bodies in the verification set and the test set to complete the model prediction process and then verify the model accuracy.

[0056] Specifically, the accuracy of the model is verified using a test set or a new ASGI remote sensing image tile that is independent of the training set. The result predicted by the duckweed-type water body recognition model is the predicted attribute value, and the data label (i.e., the attribute value manually outlined and labeled) is the true value. Then, the precision (Precision), overall accuracy (OA), recall (Recall), false alarm rate (FPR), miss rate (FNR), harmonic mean of precision and recall (F1-score) and intersection of Union (IoU) are used as indicators to evaluate the accuracy of the duckweed-type water body recognition model.

[0057] Among them, the above indicators are calculated by the true positive (TP), true negative (TN), false negative (FP), and false positive (FN) in the confusion matrix, which are introduced one by one below.

[0058] The precision is expressed as the following formula (9): (9)

[0059] The overall accuracy (OA) is expressed as follows: (10)

[0060] Recall is divided into positive sample recall And the negative sample recall rate , respectively expressed as the following formula (11) and formula (12): (11)

[0061] (12)

[0062] The false alarm rate (FPR) is expressed as follows: (13)

[0063] The missed detection rate (FNR) is expressed as the following formula (14): (14)

[0064] The F1-score is expressed as the following formula (15): (15)

[0065] The intersection over union (IoU) is expressed as follows (16): (16)

[0066] Through the above indicators, a comprehensive evaluation can be made on the model performance of the duckweed type water body recognition model. The larger the indicator, the better the performance of the duckweed type water body recognition model and the higher the recognition accuracy of duckweed type water bodies. The duckweed type water body recognition model can directly distinguish duckweed type water bodies in ASGI remote sensing image tiles in actual application scenarios.

[0067] See also Figure 3 , Figure 3 The present invention provides a schematic diagram for distinguishing easily confused objects after ASGI band combination. After forming an ASGI remote sensing image tile by combining three ASGI bands, namely, the chromatic angle band (a), the slope and band (S), and the green channel index band, the duckweed type water body recognition model is used to recognize the duckweed type water body in the ASGI remote sensing image tile. Figure 3As shown in (a) in the figure, the duckweed water body recognition model accurately identifies the duckweed water body drawn manually in the ASGI remote sensing image tile and displays the corresponding local magnified image. Figure 3 As shown in (b) in the figure, the duckweed-type water body recognition model also accurately distinguishes light green vegetation, bright blue paint, bright green paint, and red paint, and displays the corresponding local magnified images. Figure 3 As shown in (c), the duckweed-type water body recognition model can also accurately identify the most confusing dark green paint, and displays the corresponding local enlarged image.

[0068] After the duckweed type water body recognition model is verified for accuracy, it can be put into actual application environment. By collecting remote sensing tiles of unknown duckweed type water bodies, the duckweed type water body recognition model is called to effectively identify various unknown duckweed type water bodies in remote sensing tiles.

[0069] In summary, the embodiment of the present invention, based on the remote sensing tile of duckweed-type water body, constructs three physical features for identifying duckweed-type water body, namely, chromaticity angle band, slope and band, and green channel index band, and combines the three physical features to obtain ASGI remote sensing image tiles, and constructs a training data set for duckweed-type water body by setting attribute thresholds to train the initial deep learning model, obtain a duckweed-type water body recognition model, to identify duckweed-type water body, and verify the accuracy of the duckweed-type water body recognition model through various indicators during the training process, thereby ensuring the effectiveness of the duckweed-type water body recognition model. Thus, duckweed-type water body can be identified simultaneously through the three physical features of chromaticity angle band, slope and band, and green channel index band, which can eliminate the confusing effects of other objects in a targeted manner, improve the recognition accuracy, and overcome the defect of low resolution of remote sensing images in the prior art that reduces the recognition accuracy.

[0070] The duckweed type water body remote sensing monitoring device based on deep learning provided by the present invention is described below. The duckweed type water body remote sensing monitoring device based on deep learning described below and the duckweed type water body remote sensing monitoring method based on deep learning described above can be referenced to each other.

[0071] See also Figure 4 , Figure 4 is a structural schematic diagram of a duckweed-type water body remote sensing monitoring device based on deep learning provided by the present invention, such as Figure 4As shown, the duckweed-type water body remote sensing monitoring device based on deep learning includes an acquisition module 401, a determination module 402, a synthesis module 403, a construction module 404, and an identification module 405. Among them, the acquisition module 401 is used to obtain the remote sensing tile of the duckweed type water body as the RGB three-channel sample; the determination module 402 is used to perform calculation based on the chromaticity angle on the RGB three-channel sample to obtain the chromaticity angle band, perform calculation based on the slope sum on the RGB three-channel sample to obtain the slope sum band, and perform calculation based on the green channel index on the RGB three-channel sample to obtain the green channel index band; the synthesis module 403 is used to perform band synthesis on the chromaticity angle band, the slope sum band and the green channel index band to obtain the ASGI remote sensing image tile; the construction module 404 is used to set the attribute threshold according to the duckweed type water body in the ASGI remote sensing image tile, and construct a training data set for the duckweed type water body; the identification module 405 is used to train the initial deep learning model through the training data set to obtain a duckweed type water body recognition model to identify the duckweed type water body.

[0072] It can be seen from this that the beneficial effects produced by the duckweed type water body remote sensing monitoring device based on deep learning provided by the present invention should be the same as the duckweed type water body remote sensing monitoring method based on deep learning described above. Therefore, the beneficial effects produced by the duckweed type water body remote sensing monitoring device based on deep learning will not be repeated here.

[0073] Figure 5 is a schematic diagram of the physical structure of an electronic device provided by the present invention, such as Figure 5As shown, the electronic device may include: a processor (processor) 510, a communication interface (Communications Interface) 520, a memory (memory) 530 and a communication bus 840, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. Processing 510 can call the logic instructions in the memory 530 to execute a remote sensing monitoring method for duckweed-type water bodies based on deep learning, the method including: obtaining remote sensing tiles of duckweed-type water bodies as RGB three-channel samples; performing calculations based on chromaticity angles on the RGB three-channel samples to obtain chromaticity angle bands, performing calculations based on slope sums on the RGB three-channel samples to obtain slope sum bands, and performing calculations based on green channel indexes on the RGB three-channel samples to obtain green channel index bands; performing band synthesis on the chromaticity angle bands, the slope sum bands, and the green channel index bands to obtain ASGI remote sensing image tiles; setting attribute thresholds according to the duckweed-type water bodies in the ASGI remote sensing image tiles, and constructing a sample data set of duckweed-type water bodies; training an initial deep learning model through the sample data set to obtain a duckweed-type water body recognition model to identify duckweed-type water bodies.

[0074] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0075] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the deep learning-based duckweed water body remote sensing monitoring method provided by the above methods, the method including: obtaining a remote sensing tile of a duckweed water body as an RGB three-channel sample; performing a calculation based on a chromatic angle on the RGB three-channel sample to obtain a chromatic angle band, performing a calculation based on a slope sum on the RGB three-channel sample to obtain a slope sum band, and performing a calculation based on a green channel index on the RGB three-channel sample to obtain a green channel index band; performing band synthesis on the chromatic angle band, the slope sum band and the green channel index band to obtain an ASGI remote sensing image tile; setting an attribute threshold according to the duckweed water body in the ASGI remote sensing image tile, and constructing a sample data set of duckweed water bodies; training an initial deep learning model through the sample data set to obtain a duckweed water body recognition model to identify duckweed water bodies.

[0076] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the deep learning-based duckweed water body remote sensing monitoring method provided by the above-mentioned methods, the method comprising: obtaining a remote sensing tile of a duckweed water body as an RGB three-channel sample; performing a calculation based on a chromaticity angle on the RGB three-channel sample to obtain a chromaticity angle band, performing a calculation based on a slope sum on the RGB three-channel sample to obtain a slope sum band, and performing a calculation based on a green channel index on the RGB three-channel sample to obtain a green channel index band; performing band synthesis on the chromaticity angle band, the slope sum band and the green channel index band to obtain an ASGI remote sensing image tile; setting an attribute threshold according to the duckweed water body in the ASGI remote sensing image tile, and constructing a sample data set of duckweed water bodies; training an initial deep learning model through the sample data set to obtain a duckweed water body recognition model to identify duckweed water bodies.

[0077] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0078] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0079] 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 make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for remote sensing monitoring of duckweed-type water bodies based on deep learning, characterized in that: The method comprises: Get the remote sensing tiles of duckweed-type water bodies as RGB three-channel samples; Performing calculation based on the chromaticity angle on the RGB three-channel samples to obtain a chromaticity angle band, performing calculation based on the slope sum on the RGB three-channel samples to obtain a slope sum band, and performing calculation based on the green channel index on the RGB three-channel samples to obtain a green channel index band; Performing band synthesis on the chromaticity angle band, the slope sum band and the green channel index band to obtain an ASGI remote sensing image tile; Setting attribute thresholds according to the duckweed-type water bodies in the ASGI remote sensing image tiles, and constructing a sample data set of duckweed-type water bodies; The initial deep learning model is trained by the sample data set to obtain a duckweed-type water body recognition model to identify duckweed-type water bodies.

2. The method for remote sensing monitoring of duckweed-type water bodies based on deep learning according to claim 1 is characterized in that: The calculating based on the chromaticity angle of the RGB three-channel samples to obtain the chromaticity angle band includes: Get the band linear summation coefficient of map remote sensing; According to the band linear summation coefficient, the R value, the G value and the B value of the RGB three-channel sample are subjected to stimulus value calculation to obtain the XYZ three primary color stimulus values; Based on the XYZ three primary color stimulus values, determine the chromaticity abscissa and the chromaticity ordinate; The chromaticity angle is calculated for the chromaticity horizontal coordinate and the chromaticity vertical coordinate to obtain a chromaticity angle band.

3. The method for remote sensing monitoring of duckweed-type water bodies based on deep learning according to claim 1 is characterized in that: The step of performing a slope-based sum calculation on the RGB three-channel samples to obtain a slope and a band includes: Determine a first ratio of the B value to the G value in the RGB three-channel sample; Determine a second ratio of the R value to the G value in the RGB three-channel sample; The sum of the first ratio and the second ratio is taken as the slope and band.

4. The method for remote sensing monitoring of duckweed-type water bodies based on deep learning according to claim 1, characterized in that: The green channel index band is obtained by calculating the green channel index based on the RGB three-channel samples, including: Determine the sum of the channel values ​​of the B value and the R value in the RGB three-channel sample; The ratio of the square of the G value in the RGB three-channel sample to the sum of the channel values ​​is used as the green channel index band.

5. The method for remote sensing monitoring of duckweed-type water bodies based on deep learning according to claim 1, characterized in that: The method of setting an attribute threshold according to the duckweed type water body in the ASGI remote sensing image tile and constructing a sample data set of the duckweed type water body includes: Determine the surface vector of the duckweed-type water body in the ASGI remote sensing image tile; When the duckweed-type water body is a real duckweed-type water body, setting a first attribute threshold for the surface vector of the duckweed-type water body; When the duckweed type water body is a non-duckweed type water body, setting a second attribute threshold for the surface vector of the duckweed type water body; The data label corresponding to the ASGI remote sensing image tile is constructed according to the first attribute threshold and the second attribute threshold, and a sample data set of duckweed-type water bodies is obtained.

6. The method for remote sensing monitoring of duckweed-type water bodies based on deep learning according to claim 1, characterized in that: The initial deep learning model is trained by the sample data set to obtain a duckweed type water body recognition model, including: Inputting the training samples of the sample data set into the initial deep learning model for forward propagation to obtain the predicted attribute values ​​of the training samples for duckweed-type water bodies; Constructing a training loss value according to the difference between the predicted attribute value and the data label of the training sample, and back-propagating the training loss value in the initial deep learning model to update the model parameters of the initial deep learning model; When the training loss value begins to converge or reaches a preset training round, the training of the initial deep learning model is terminated to obtain a duckweed-type water body recognition model.

7. A duckweed-type water body remote sensing monitoring device based on deep learning, characterized in that: The device comprises: The acquisition module is used to obtain the remote sensing tiles of duckweed-type water bodies as RGB three-channel samples; A determination module, configured to perform a calculation based on a chromaticity angle on the RGB three-channel samples to obtain a chromaticity angle band, perform a calculation based on a slope sum on the RGB three-channel samples to obtain a slope sum band, and perform a calculation based on a green channel index on the RGB three-channel samples to obtain a green channel index band; A synthesis module, used for synthesizing the chromaticity angle band, the slope sum band and the green channel index band to obtain an ASGI remote sensing image tile; A construction module, used to set attribute thresholds according to the duckweed-type water bodies in the ASGI remote sensing image tiles, and to construct a training data set for the duckweed-type water bodies; The recognition module is used to train the initial deep learning model through the training data set to obtain a duckweed-type water body recognition model to identify duckweed-type water bodies.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the deep learning-based remote sensing monitoring method for duckweed-type water bodies is implemented as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the deep learning-based remote sensing monitoring method for duckweed-type water bodies as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the deep learning-based remote sensing monitoring method for duckweed-type water bodies as described in any one of claims 1 to 6 is implemented.

Images