CNN-based AUV underwater docking depth fuzzy sensing method

Abstract

The invention relates to the technical field of depth perception, and discloses a CNN-based AUV underwater docking depth fuzzy perception method, which comprises the following steps: constructing a visual marker composed of a two-color laser line light source, and arranging the visual marker in a docking AUV; the method comprises the following steps: shooting visual markers of a docking AUV at different distances through a camera carried by a target AUV to obtain a plurality of sample images, and formulating a depth fuzzy coding rule based on the depth difference between the target AUV and the docking AUV and the color of the visual markers in the sample images; labeling category labels for the sample images based on a depth fuzzy coding rule to obtain a training sample set; constructing a depth fuzzy perception model based on the CNN, and performing iterative training on the depth fuzzy perception model based on the training sample set until convergence to obtain a trained model; and inputting a to-be-recognized image into the trained model to obtain a depth fuzzy perception result of the to-be-recognized image. According to the method, the underwater docking efficiency of the AUV is effectively improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of depth perception in underwater docking, and particularly to a method for depth fuzzy perception of AUV underwater docking based on CNN. Background Art

[0002] With the rapid development of deep-sea technology, how to achieve energy and information transfer has become a research hotspot. Autonomous Underwater Vehicle (AUV) has the characteristics of small volume and high flexibility, but the energy it carries is limited. The common energy replenishment and information transfer are achieved through the recovery of the surface mother ship, which is time-consuming and dependent on the mother ship.

[0003] In the face of this situation, the currently common solution is to use an underwater dock (underwater docking point) for wireless charging and information transfer of AUV. The key step in this process is to sense the relative position information between the AUV and the underwater dock. The commonly used method is to use a point light source matrix as a visual marker. However, limited by the angle of the head camera in the vertical direction, the perception range in the vertical direction is severely limited in actual use. Therefore, it is necessary to search at different depths until the point light source marker is found. Therefore, the essence of this method is an exhaustive method, which takes a long time and results in poor efficiency of AUV underwater docking.

[0004] How to shorten the search time and improve the efficiency of AUV underwater docking is a technical problem that urgently needs to be solved. At present, some research teams have proposed adding cameras to AUV to supplement the perception in the vertical direction. However, in order to achieve the perception in the upper and lower directions, at least two cameras need to be added. This method not only greatly increases the computational workload of image processing, but also needs to solve the coupling problem between different cameras, and the power consumption will also increase. Summary of the Invention

[0005] In order to solve the above technical problems, the embodiments of the present application propose a method for depth fuzzy perception of AUV underwater docking based on CNN, which can improve the depth perception ability of AUV in the vertical direction during underwater docking, thereby reducing the time of AUV underwater docking and improving the efficiency of AUV underwater docking.

[0006] To achieve the above object, an embodiment of the present application proposes a depth fuzzy perception method for AUV underwater docking based on CNN. The method includes the following steps: constructing a visual marker composed of a bicolor laser line light source and setting the constructed visual marker inside the docking AUV; wherein, the bicolor laser line light source is composed of four line lasers with a common emission point including two green lasers and two red lasers, the two green lasers are distributed on one side, and the two red lasers are distributed on the other side; taking a plurality of sample images by a camera carried by the target AUV facing the visual marker of the docking AUV at different distances, and formulating a depth fuzzy coding rule based on the depth difference between the target AUV and the docking AUV and the color of the visual marker in the sample images; dividing the plurality of sample images into a plurality of different categories based on the depth fuzzy coding rule, labeling category labels for each sample image to obtain a training sample set; constructing a depth fuzzy perception model based on CNN, inputting the sample images in the training sample set into the depth fuzzy perception model for iterative training until convergence to obtain a trained depth fuzzy perception model; inputting the image to be recognized into the trained depth fuzzy perception model to obtain the depth fuzzy perception result of the trained depth fuzzy perception model for the image to be recognized, and the image to be recognized is obtained by the target AUV taking a picture of the visual marker of the docking AUV.

[0007] To achieve the above object, an embodiment of the present application further provides an AUV underwater docking depth fuzzy perception system based on CNN. The system includes: a visual marker construction module, a depth fuzzy coding rule formulation module, a training sample set construction module, a model construction module, a model training module, and a model usage module; The visual marker construction module is used to construct a visual marker composed of a two-color laser line light source and set the constructed visual marker inside the docking AUV. Among them, the two-color laser line light source is composed of four co-emission point line lasers including two green lasers and two red lasers. The two green lasers are distributed on one side, and the two red lasers are distributed on the other side; The depth fuzzy coding rule formulation module is used to take pictures of the visual marker of the docking AUV at different distances through the camera carried by the target AUV to obtain a number of sample images, and formulate depth fuzzy coding rules based on the depth difference between the target AUV and the docking AUV and the color of the visual marker in the sample images; The training sample set construction module is used to divide a number of sample images into several different categories based on the depth fuzzy coding rules, label category labels for each sample image to obtain a training sample set; The model construction module is used to construct a depth fuzzy perception model based on CNN; The model training module is used to input the sample images in the training sample set into the depth fuzzy perception model for iterative training until convergence to obtain a trained depth fuzzy perception model; The model usage module is used to input the image to be recognized into the trained depth fuzzy perception model to obtain the depth fuzzy perception result of the trained depth fuzzy perception model for the image to be recognized. The image to be recognized is obtained by the target AUV taking pictures of the visual marker of the docking AUV.

[0008] Correspondingly, an embodiment of the present application further provides an electronic device. The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute an AUV underwater docking depth fuzzy perception method based on CNN as described above.

[0009] Correspondingly, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which when executed by a processor, can implement the AUV underwater docking depth fuzzy perception method based on CNN as described above.

[0010] An AUV underwater docking depth fuzzy perception method proposed by an embodiment of the present application. Considering the problem that the perception range of an AUV in the depth direction is small during underwater docking due to the limitation of the camera's vertical direction angle and the light-emitting range of the LED point light source marker, a visual marker composed of a dual-color laser line light source, a depth fuzzy coding rule, and a depth fuzzy perception model constructed based on CNN are designed to improve the AUV's depth perception ability in the vertical direction during underwater docking. The dual-color laser line light source consists of four co-emitting point line lasers including two green lasers and two red lasers. The two green lasers are distributed on one side, and the two red lasers are distributed on the other side. This mixed form of red laser and green laser can well solve the direction conflict in depth perception. The purpose of depth fuzzy perception is not to obtain the specific value of the depth, but to perform fuzzy coding on the depth and then decode it into corresponding categories to obtain a general depth perception. Therefore, based on the depth difference between the target AUV and the docking AUV and the color of the visual marker in the sample image, the present application formulates a depth fuzzy coding rule, thereby fuzzily coding the relative position between the target AUV and the docking AUV into several different categories. Finally, by combining the depth fuzzy coding with CNN, the perception in the depth direction is realized through the feature extraction of the image to be recognized, which well meets the requirements of underwater docking, reduces the time of AUV underwater docking, and improves the efficiency of AUV underwater docking.

[0011] In some alternative embodiments, in the dual-color laser line light source, two red lasers are located on the left side, and two green lasers are located on the right side, and the included angle between two adjacent lasers is 90 degrees.

[0012] In some alternative embodiments, through the camera carried by the target AUV, the visual marker of the docking AUV is photographed at different distances to obtain a number of sample images, including: indicating that the target AUV approaches the docking AUV. When it is detected that the distance between the target AUV and the docking AUV is less than 20 meters, the camera carried by the target AUV is used to photograph the visual marker of the docking AUV; indicating that the target AUV approaches the docking AUV at a close step of 1 meter. The camera carried by the target AUV is used to photograph the visual marker of the docking AUV at different distances until the distance between the target AUV and the docking AUV is less than 1 meter, thereby obtaining a number of sample images. By photographing at different distances, sample images in different situations can be obtained, thus providing rich training data for subsequent model training.

[0013] In some alternative embodiments, based on the depth difference between the target AUV and the docking AUV and the color of the visual marker in the sample image, a depth blur coding rule is formulated, including: if the absolute value of the depth difference between the target AUV and the docking AUV is less than or equal to the first preset threshold, it is considered that the target AUV is in the middle front of the docking AUV; if the depth difference between the target AUV and the docking AUV is greater than the first preset threshold, it is considered that the target AUV is in the upper front position of the docking AUV; if the depth difference between the target AUV and the docking AUV is less than the second preset threshold, it is considered that the target AUV is in the lower front position of the docking AUV; wherein, the first preset threshold and the second preset threshold are opposite numbers to each other; if the difference between red and green in the color of the visual marker in the sample image is less than or equal to the third preset threshold, it is considered that the target AUV is in the middle front of the docking AUV; if the color of the visual marker in the sample image is biased towards red and the difference between red and green is greater than the third preset threshold, it is considered that the target AUV is in the upper right front position of the docking AUV; if the color of the visual marker in the sample image is biased towards green and the difference between red and green is greater than the third preset threshold, it is considered that the target AUV is in the lower left front position of the docking AUV. The depth difference between the target AUV and the docking AUV, as well as the color of the visual marker in the sample image, can well represent the relative position between the target AUV and the docking AUV, thereby blurring the specific depth information into azimuth information and enhancing the depth perception ability in the vertical direction.

[0014] In some alternative embodiments, based on the depth blur coding rule, a number of sample images are divided into several different categories, and class labels are assigned to each sample image to obtain a training sample set, including: based on the depth blur coding rule, a number of sample images are divided into 9 different categories, namely upper left, upper, upper right, left middle, middle, right middle, lower left, lower, and lower right, which respectively represent that the target AUV is in the upper left front position, the upper front position directly ahead, the upper right front position, the left front position directly ahead, the front position directly ahead, the right front position directly ahead, the lower left front position, the lower front position directly ahead, and the lower right front position of the docking AUV; use to represent the class label of the sample image, wherein, use to characterize upper left, use to characterize upper, use to characterize upper right, use to characterize left middle, use to characterize middle, use to characterize right middle, use to characterize lower left, use to characterize lower, use to characterize lower right.

[0015] In some alternative embodiments, the deep blur perception model based on CNN consists of an input layer, a hidden layer, and an output layer; the input layer is used to extract color features and grayscale features from the sample images; the hidden layer consists of a pooling layer, a convolutional layer, and a fully connected layer; the output layer is used to perform blur encoding and depth decoding based on the output of the hidden layer, and output the deep blur perception result; when iteratively training the deep blur perception model, the gradient descent method is used to optimize the model parameters of the deep blur perception model. The deep blur perception model not only extracts the color features in the image, but also needs to extract the grayscale features, and then through the careful processing of the hidden layer, so as to accurately output the deep blur perception result.

[0016] In some alternative embodiments, when the target AUV takes pictures of the visual marker of the docking AUV, N pictures are continuously taken as the images to be recognized, where N is an integer greater than 1; when inputting the images to be recognized into the trained deep blur perception model, the N pictures are input into the trained deep blur perception model at the same time, and the trained deep blur perception model performs mean filtering on the N pictures and then stitches them into one picture as the input to be sent to the input layer. The design of mean filtering can increase the credibility of the output deep blur perception result. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the related art, the following will briefly introduce the drawings required for the description of the embodiments of the present application or the related art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0018] Figure 1 is a flowchart of a method for deep blur perception of AUV underwater docking based on CNN provided in an embodiment of the present application;

[0019] Figure 2 is a schematic diagram of a dual-color laser line light source provided in an embodiment of the present application;

[0020] Figure 3 is a schematic diagram of the corresponding relationship of the deep blur encoding rules provided in an embodiment of the present application;

[0021] Figure 4 is a schematic diagram of the position of the deep blur encoding rules in space provided in an embodiment of the present application;

[0022] Figure 5 is a schematic diagram of constructing a training sample set provided in an embodiment of the present application;

[0023] Figure 6It is a schematic structural diagram of a depth blur perception model provided in an embodiment of the present application;

[0024] Figure 7 It is a schematic diagram of the use of the model provided in an embodiment of the present application;

[0025] Figure 8 It is a schematic structural diagram of an AUV underwater docking depth blur perception system based on CNN provided in another embodiment of the present application;

[0026] Figure 9 It is a schematic structural diagram of an electronic device provided in another embodiment of the present application. Detailed implementation manners

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the embodiments of the present application will be elaborated in detail below with reference to the accompanying drawings. In various embodiments of the present application, many technical details are proposed to help readers better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solutions claimed in the present application can still be implemented. The division of the following embodiments is only for convenient description and should not constitute any limitation on the specific implementation manners of the present application. The various embodiments can be combined and cross-referenced with each other on the premise of not being contradictory.

[0028] An embodiment of the present application proposes a CNN-based AUV underwater docking depth blur perception method, which is applied to an electronic device. The implementation details of the CNN-based AUV underwater docking depth blur perception method proposed in this embodiment will be specifically described below. The following content is only implementation details provided for convenient understanding and is not necessary for implementing this solution.

[0029] The specific process of the CNN-based AUV underwater docking depth blur perception method proposed in this embodiment can be as Figure 1 shown and includes:

[0030] Step 101, construct a visual marker composed of a two-color laser line light source, and set the constructed visual marker inside the docking AUV. Among them, the two-color laser line light source is composed of four line lasers with a common emission point, including two green lasers and two red lasers. The two green lasers are distributed on one side, and the two red lasers are distributed on the other side.

[0031] In a specific implementation, due to the limited light-emitting range of a point light source, in this embodiment, a visual marker composed of a dual-color laser line light source is constructed. The visual marker has high penetration ability and can be arranged inside the docking AUV. The dual-color laser line light source is composed of four line lasers with a common emission point, including two green lasers and two red lasers. The two green lasers are distributed on one side, and the two red lasers are distributed on the other side. In this way, the images obtained by photographing the visual marker at different positions will have color differences.

[0032] In one example, the dual-color laser line light source is as Figure 2 shown. In the dual-color laser line light source, two red lasers are located on the left side, and two green lasers are located on the right side. The included angle α between two adjacent lasers is 90 degrees. This mixing form of red laser and green laser can well solve the direction conflict in depth perception.

[0033] In one example, in the dual-color laser line light source, two red lasers are located on the right side, and two green lasers are located on the left side. The included angle α between two adjacent lasers is 90 degrees.

[0034] Step 102: Through the camera carried by the target AUV, photograph the visual marker of the docking AUV at different distances to obtain a number of sample images, and formulate a depth blur coding rule based on the depth difference between the target AUV and the docking AUV and the color of the visual marker in the sample images.

[0035] Step 103: Divide the number of sample images into several different categories based on the depth blur coding rule, label category labels for each sample image, and obtain a training sample set.

[0036] In a specific implementation, the purpose of depth blur perception is not to obtain the specific value of the depth, but to perform blur coding on the depth and then decode it into the corresponding category to obtain a general depth perception. Therefore, in this embodiment, through the camera carried by the target AUV, photograph the visual marker of the docking AUV at different distances to obtain a number of sample images, and formulate a depth blur coding rule based on the depth difference between the target AUV and the docking AUV and the color of the visual marker in the sample images. Thus, divide the number of sample images into several different categories based on the depth blur coding rule, that is, blur code the relative position between the target AUV and the docking AUV into several different categories, label category labels for each sample image, and obtain a training sample set.

[0037] In one example, when capturing sample images, the target AUV needs to be instructed to approach the docking AUV. When the distance between the target AUV and the docking AUV is detected to be less than 20 meters, the camera carried by the target AUV takes pictures facing the visual marker of the docking AUV. Subsequently, the target AUV is instructed to approach the docking AUV with an approach step of 1 meter. The camera carried by the target AUV takes pictures facing the visual marker of the docking AUV at different distances until the distance between the target AUV and the docking AUV is less than 1 meter, thereby obtaining a number of sample images. It should be noted that in order to expand the quantity and quality of the sample images, we need to take pictures in different waters. By taking pictures at different distances and in different waters, sample images in different situations can be obtained, thus providing rich training data for subsequent model training.

[0038] In one example, in addition to constructing a training sample set, a test sample set also needs to be constructed. When the distance between the target AUV and the docking AUV is not an integer, the camera carried by the target AUV takes pictures facing the visual marker of the docking AUV at different distances. The sample images obtained in this way will be classified into the test sample set.

[0039] In one example, if the absolute value of the depth difference between the target AUV and the docking AUV is less than or equal to the first preset threshold, it is considered that the target AUV is in the middle front of the docking AUV; if the depth difference between the target AUV and the docking AUV is greater than the first preset threshold, it is considered that the target AUV is in the upper front position of the docking AUV; if the depth difference between the target AUV and the docking AUV is less than the second preset threshold, it is considered that the target AUV is in the lower front position of the docking AUV. Among them, the first preset threshold and the second preset threshold are opposite to each other.

[0040] In one example, if the difference between red and green in the color of the visual marker in the sample image is less than or equal to the third preset threshold, it is considered that the target AUV is in the middle front of the docking AUV; if the color of the visual marker in the sample image is redder and the difference between red and green is greater than the third preset threshold, it is considered that the target AUV is in the upper right position of the docking AUV; if the color of the visual marker in the sample image is greener and the difference between red and green is greater than the third preset threshold, it is considered that the target AUV is in the upper left position of the docking AUV.

[0041] It can be understood that the depth difference between the target AUV and the docking AUV, as well as the color of the visual marker in the sample image, can well characterize the relative position between the target AUV and the docking AUV, thereby fuzzing the specific depth information into azimuth information and enhancing the depth perception ability in the vertical direction.

[0042] In one example, the correspondence of the depth blur encoding rules is as Figure 3 shown, and its position in space can be as Figure 4 shown. When performing classification and labeling, several sample images are divided into 9 different categories based on the depth blur encoding rules, namely upper left, upper, upper right, left middle, middle, right middle, lower left, lower, and lower right, which respectively represent the positions of the target AUV that are slightly above the left of the docking AUV, slightly above the due front, slightly above the right of the front, slightly to the left of the due front, at the due front, slightly to the right of the due front, slightly below the left of the front, slightly below the due front, and slightly below the right of the front. We use to represent the class label of the sample image. Among them, is used to represent upper left, is used to represent upper, is used to represent upper right, is used to represent left middle, is used to represent middle, is used to represent right middle, is used to represent lower left, is used to represent lower, is used to represent lower right.

[0043] In one example, the process of data collection and construction of the test sample set is as Figure 5 shown.

[0044] Step 104: Build a depth blur perception model based on CNN, input the sample images in the training sample set into the depth blur perception model for iterative training until convergence, and obtain the trained depth blur perception model.

[0045] In a specific implementation, after obtaining the training sample set, a depth blur perception model can be built based on CNN and trained. Input the sample images in the training sample set into the depth blur perception model, calculate the loss value based on the class label and the depth blur perception result output by the depth blur perception model, and perform iterative training on the depth blur perception model based on the loss value until convergence to obtain the trained depth blur perception model.

[0046] In one example, the specific structure of the depth blur perception model is as Figure 6As shown in the figure, the depth blur perception model consists of an input layer, a hidden layer, and an output layer. The input layer is used to extract color features and grayscale features from the sample image. The hidden layer consists of a pooling layer, a convolutional layer, and a fully connected layer. The output layer is used to perform blur encoding and depth decoding based on the output of the hidden layer, and output the depth blur perception result. When iteratively training the depth blur perception model, the gradient descent method can be used to optimize the model parameters. The depth blur perception model not only extracts color features in the image, but also needs to extract grayscale features, and then through the careful processing of the hidden layer, so as to accurately output the depth blur perception result.

[0047] In one example, after completing the training of the depth blur perception model, it is also necessary to use the test sample set to perform a performance test on the trained depth blur perception model. In the case of passing the performance test, the trained depth blur perception model is deployed to the required scenario, otherwise it is necessary to return to the training step and retrain.

[0048] Step 105, input the image to be recognized into the trained depth blur perception model, and obtain the depth blur perception result of the trained depth blur perception model for the image to be recognized. The image to be recognized is obtained by the target AUV taking pictures of the visual marker of the docking AUV.

[0049] In a specific implementation, after the trained depth blur perception model is deployed and put into actual use, the target AUV takes pictures of the visual marker of the docking AUV to obtain the image to be recognized, and inputs the image to be recognized into the trained depth blur perception model, then the depth blur perception result of the trained depth blur perception model for the image to be recognized can be obtained.

[0050] In one example, the usage process of the model is as Figure 7 shown. When the target AUV takes pictures of the visual marker of the docking AUV, N images are continuously taken as the images to be recognized, where N is an integer greater than 1 (usually set to 3). When inputting the images to be recognized into the trained depth blur perception model, the N images are input into the trained depth blur perception model at the same time. The trained depth blur perception model performs mean filtering on the N images and then stitches them into one image, and uses the stitched image as the input to send to the input layer. The design of mean filtering can increase the credibility of the output depth blur perception result and further improve the accuracy of depth blur perception.

[0051] An AUV underwater docking depth fuzzy perception method based on CNN is proposed in this embodiment. Considering the problem that the AUV is limited by the vertical angle of the camera and the light-emitting range of the LED point light source marker during underwater docking, and the perception range in the depth direction is small, a visual marker composed of a two-color laser line light source, a depth fuzzy coding rule, and a depth fuzzy perception model constructed based on CNN are designed to improve the AUV's depth perception ability in the vertical direction during underwater docking. The two-color laser line light source consists of four co-emitting point line lasers including two green lasers and two red lasers. The two green lasers are distributed on one side, and the two red lasers are distributed on the other side. This mixed form of red laser and green laser can well solve the direction conflict in depth perception. The purpose of depth fuzzy perception is not to obtain the specific value of the depth, but to perform fuzzy coding on the depth and then decode it into corresponding categories to obtain a rough depth perception. Therefore, in this embodiment, based on the depth difference between the target AUV and the docking AUV and the color of the visual marker in the sample image, a depth fuzzy coding rule is formulated, so as to fuzzy code the relative position between the target AUV and the docking AUV into several different categories. Finally, by combining the depth fuzzy coding with CNN, the perception in the depth direction is realized through feature extraction of the image to be recognized, which well meets the requirements of underwater docking, reduces the time of AUV underwater docking, and improves the efficiency of AUV underwater docking.

[0052] The step division of the above various methods is only for clear description. When implemented, they can be combined into one step or some steps can be split into multiple steps. As long as the same logical relationship is included, it is within the protection scope of this application; adding insignificant modifications to the algorithm or process or introducing insignificant designs, but not changing the core design of its algorithm and process are within the protection scope of this application.

[0053] Another embodiment of this application proposes an AUV underwater docking depth fuzzy perception system based on CNN. The implementation details of an AUV underwater docking depth fuzzy perception system based on CNN proposed in this embodiment will be specifically described below. The following content is only implementation details provided for convenient understanding and is not necessary for implementing this example.

[0054] Figure 8 It is a schematic structural diagram of an AUV underwater docking depth fuzzy perception system based on CNN proposed in this embodiment. The system includes: a visual marker construction module 201, a depth fuzzy coding rule formulation module 202, a training sample set construction module 203, a model construction module 204, a model training module 205, and a model usage module 206.

[0055] The visual marker construction module 201 is used to construct a visual marker composed of a dual-color laser line light source and set the constructed visual marker inside the docking AUV. The dual-color laser line light source is composed of four line lasers with a common emission point, including two green lasers and two red lasers. The two green lasers are distributed on one side, and the two red lasers are distributed on the other side.

[0056] The depth blur coding rule formulation module 202 is used to take pictures of the visual marker of the docking AUV at different distances through the camera carried by the target AUV to obtain a number of sample images, and formulate a depth blur coding rule based on the depth difference between the target AUV and the docking AUV and the color of the visual marker in the sample images.

[0057] The training sample set construction module 203 is used to divide a number of sample images into several different categories based on the depth blur coding rule, label category labels for each sample image, and obtain a training sample set.

[0058] The model construction module 204 is used to construct a depth blur perception model based on CNN.

[0059] The model training module 205 is used to input the sample images in the training sample set into the depth blur perception model for iterative training until convergence, and obtain a trained depth blur perception model.

[0060] The model usage module 206 is used to input the image to be recognized into the trained depth blur perception model, and obtain the depth blur perception result of the trained depth blur perception model for the image to be recognized. The image to be recognized is obtained by the target AUV taking pictures of the visual marker of the docking AUV.

[0061] It is worth mentioning that each module involved in this embodiment is a logical module. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, to highlight the innovative part of this application, units that are not closely related to solving the technical problems proposed in this application are not introduced in this embodiment, but this does not mean that there are no other units in this embodiment.

[0062] It is not difficult to find that this embodiment is a system embodiment corresponding to the above method embodiment. This embodiment can be implemented in cooperation with the above method embodiment. The relevant technical details and technical effects mentioned in the above method embodiment are still valid in this embodiment. To avoid repetition, they are not elaborated here. Correspondingly, the relevant technical details mentioned in this embodiment can also be applied in the above method embodiment.

[0063] Another embodiment of the present application proposes an electronic device, and its specific structure is as follows Figure 9 shown, including: at least one processor 301; and a memory 302 communicatively connected to the at least one processor 301; wherein, the memory 302 stores instructions executable by the at least one processor 301, and the instructions are executed by the at least one processor 301 to enable the at least one processor 301 to execute a method for depth fuzzy perception of AUV underwater docking based on CNN as described in the above method embodiment.

[0064] Among them, the memory and the processor can be connected in a bus manner. The bus can include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and the memory together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits together, which are well known in the art and will not be further described herein. The bus interface is responsible for providing an interface between the bus and the transceiver. The transceiver can be one component or multiple components, such as multiple receivers and transmitters, and provides a unit for communicating with various other devices on the transmission medium.

[0065] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interface, voltage regulation, power management, and other control functions. And the memory can be used to store data used by the processor when executing operations.

[0066] Another embodiment of the present application proposes a computer-readable storage medium storing a computer program, which when executed by a processor, can implement a method for depth fuzzy perception of AUV underwater docking based on CNN as described in the above method embodiment.

[0067] That is, those skilled in the art can understand that all or part of the steps in implementing the above method embodiments can be completed by instructing relevant hardware through a program. This program is stored in a storage medium and includes several instructions to enable a device (such as a single-chip microcomputer, a chip, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. And the foregoing storage medium includes: USB flash drives, mobile hard disks, ROM (Read-Only Memory), RAM (Random Access Memory), magnetic disks, or optical disks and other various media that can store program codes.

[0068] Those of ordinary skill in the art can understand that the above embodiments are specific embodiments for implementing the present application, and in practical applications, various changes can be made in form and details without departing from the spirit and scope of the present application.

Claims

1. An AUV underwater docking depth fuzzy perception method based on CNN, characterized in that, Including: Construct a visual marker composed of a two-color laser line light source, and set the constructed visual marker inside the docking AUV; wherein, the two-color laser line light source is composed of four line lasers with a common emission point including two green lasers and two red lasers, the two green lasers are distributed on one side, and the two red lasers are distributed on the other side; Through the camera carried by the target AUV, take pictures of the visual marker of the docking AUV at different distances to obtain a number of sample images, and formulate a depth blur coding rule based on the depth difference between the target AUV and the docking AUV and the color of the visual marker in the sample images; Based on the depth blur coding rule, divide a number of sample images into a number of different categories, label category labels for each sample image to obtain a training sample set; Construct a depth blur perception model based on CNN, input the sample images in the training sample set into the depth blur perception model for iterative training until convergence to obtain a trained depth blur perception model; Input the image to be recognized into the trained depth blur perception model to obtain the depth blur perception result of the trained depth blur perception model for the image to be recognized, and the image to be recognized is obtained by the target AUV taking pictures of the visual marker of the docking AUV.

2. The AUV underwater docking depth fuzzy perception method based on CNN according to claim 1, characterized in that, In the two-color laser line light source, the two red lasers are on the left side, and the two green lasers are on the right side, and the included angle between two adjacent lasers is 90 degrees.

3. A method for underwater docking depth fuzzy perception of an AUV based on CNN according to claim 2, characterized in that, The step of taking pictures of the visual marker of the docking AUV at different distances through the camera carried by the target AUV to obtain a number of sample images includes: Instruct the target AUV to approach the docking AUV. When it is detected that the distance between the target AUV and the docking AUV is less than 20 meters, use the camera carried by the target AUV to take pictures of the visual marker of the docking AUV; Instruct the target AUV to approach the docking AUV at an approaching step size of 1 meter. Use the camera carried by the target AUV to take pictures of the visual marker of the docking AUV at different distances until the distance between the target AUV and the docking AUV is less than 1 meter, so as to obtain a number of sample images.

4. The AUV underwater docking depth fuzzy perception method based on CNN according to claim 3, characterized in that, Formulate a depth blur coding rule based on the depth difference between the target AUV and the docking AUV and the color of the visual marker in the sample images, including: If the absolute value of the depth difference between the target AUV and the docking AUV is less than or equal to the first preset threshold, it is considered that the target AUV is in the middle front of the docking AUV. If the depth difference between the target AUV and the docking AUV is greater than the first preset threshold, it is considered that the target AUV is in the upper front position of the docking AUV. If the depth difference between the target AUV and the docking AUV is less than the second preset threshold, it is considered that the target AUV is in the lower front position of the docking AUV; wherein, the first preset threshold and the second preset threshold are opposite to each other; If the difference between red and green in the color of the visual marker in the sample image is less than or equal to the third preset threshold, it is considered that the target AUV is in the middle front of the docking AUV. If the color of the visual marker in the sample image is red-biased and the difference between red and green is greater than the third preset threshold, it is considered that the target AUV is in the position slightly to the right front of the docking AUV. If the color of the visual marker in the sample image is green-biased and the difference between red and green is greater than the third preset threshold, it is considered that the target AUV is in the position slightly to the left front of the docking AUV.

5. A method for fuzzy perception of AUV underwater docking depth based on CNN according to claim 4, characterized in that, Based on the depth fuzzy coding rule, several sample images are divided into several different categories, and category labels are assigned to each sample image to obtain a training sample set, including: Based on the depth fuzzy coding rule, several sample images are divided into 9 different categories, namely upper left, upper, upper right, left middle, middle, right middle, lower left, lower, and lower right, which respectively represent the position of the target AUV slightly to the upper left front, directly above the front, slightly to the upper right front, directly to the left front, directly in front, directly to the right front, slightly to the lower left front, directly below the front, and slightly to the lower right front of the docking AUV; Use to represent the class label of the sample image, where characterizes the upper left, characterizes the upper, characterizes the upper right, characterizes the middle left, characterizes the middle, characterizes the middle right, characterizes the lower left, characterizes the lower, characterizes the lower right.

6. A method for depth fuzzy perception of AUV underwater docking based on CNN according to claim 1, characterized in that, The depth fuzzy perception model constructed based on CNN consists of an input layer, a hidden layer, and an output layer; The input layer is used to extract color features and grayscale features from the sample image; The hidden layer consists of a pooling layer, a convolutional layer, and a fully connected layer; The output layer is used to perform fuzzy coding and depth decoding based on the output of the hidden layer and output the depth fuzzy perception result; When performing iterative training on the depth fuzzy perception model, the gradient descent method is used to optimize the model parameters.

7. A method for depth fuzzy perception of AUV underwater docking based on CNN according to claim 6, characterized in that When the target AUV takes pictures of the visual marker facing the docking AUV, N images are continuously taken as the images to be recognized, where N is an integer greater than 1; When inputting the images to be recognized into the trained depth fuzzy perception model, the N images are input into the trained depth fuzzy perception model at the same time. The trained depth fuzzy perception model performs mean filtering on the N images and then stitches them into one image, which is sent to the input layer as the input.

8. An AUV underwater docking depth fuzzy perception system based on CNN, characterized in that, Including: A visual marker construction module, a depth fuzzy coding rule formulation module, a training sample set construction module, a model construction module, a model training module, and a model usage module; The visual marker construction module is used to construct a visual marker composed of a two-color laser line light source and set the constructed visual marker inside the docking AUV. Among them, the two-color laser line light source consists of four line lasers with a common emission point, including two green lasers and two red lasers. The two green lasers are distributed on one side, and the two red lasers are distributed on the other side; The depth fuzzy coding rule formulation module is used to take pictures of the visual marker of the docking AUV at different distances through the camera carried by the target AUV to obtain several sample images, and formulate the depth fuzzy coding rule based on the depth difference between the target AUV and the docking AUV and the color of the visual marker in the sample image; A training sample set construction module, configured to classify a plurality of sample images into a plurality of different categories based on deep fuzzy coding rules, label category labels for each sample image, and obtain a training sample set; A model construction module, configured to construct a deep fuzzy perception model based on CNN; A model training module, configured to input the sample images in the training sample set into the deep fuzzy perception model for iterative training until convergence, and obtain a trained deep fuzzy perception model; A model usage module, configured to input the image to be recognized into the trained deep fuzzy perception model, and obtain the deep fuzzy perception result of the trained deep fuzzy perception model for the image to be recognized, where the image to be recognized is obtained by the target AUV taking a picture of the visual marker of the docking AUV.

9. An electronic device, characterized in that, Including: At least one processor; And a memory communicatively connected to the at least one processor; Wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor, so that the at least one processor can execute a CNN-based AUV underwater docking deep fuzzy perception method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it can implement a CNN-based AUV underwater docking deep fuzzy perception method according to any one of claims 1 to 7.

Images