Disclosure of Invention
The application solves the problem of how to extract the submarine landform features deeply in the submarine landform classification process, and improves the characterization capability of the landform features.
In order to solve the problems, the application provides a method for constructing an intelligent classification model of submarine topography, which comprises the following steps:
acquiring multi-beam data of an original sea area;
preprocessing the multi-beam data of the original sea area to obtain original back scattering data and original landform class label data;
training and optimizing an original classification model according to the original back scattering data and the original landform class label data to obtain a submarine landform classification model; the original classification model is constructed based on an improved label smooth cross entropy loss function and a pruning type dense connection network, and the submarine topography classification model is used for predicting submarine topography categories of sea areas.
Optionally, the construction process of the original classification model includes:
acquiring an original dense connection network;
pruning operation is carried out on the original dense connection network, and the pruning dense connection network is obtained based on a downsampling function;
and obtaining the original classification model according to the improved label smooth cross entropy loss function and the pruning type dense connection network.
Optionally, training and optimizing the original classification model according to the original backscatter data and the original landform class label data to obtain a submarine landform classification model, including:
inputting the original backscatter data into the original classification model for training to obtain a temporary prediction classification result;
and performing tuning operation according to the temporary prediction classification result and the original landform class label data, and taking the tuned original classification model as the submarine landform classification model.
Optionally, the optimizing operation is performed according to the temporary prediction classification result and the original landform class label data, and the optimized original classification model is used as the submarine landform classification model, which includes:
performing loss calculation through the improved label smooth cross entropy loss function according to the temporary prediction classification result and the original landform class label data to obtain loss function output;
model parameters of the original classification model are adjusted according to the loss function output until the loss function input meets preset conditions, and the original classification model after parameter adjustment is used as the submarine topography classification model;
wherein the improved label smoothing cross entropy loss function is obtained by a label smoothing function and a cross entropy loss function.
Optionally, the improved label smoothing cross entropy loss function is obtained by a label smoothing function and a cross entropy loss function, comprising:
obtaining the improved label smoothing cross entropy loss function by carrying out fusion operation on the label smoothing function and the cross entropy loss function; wherein the tag smoothing function is:
y′=(1-∈)y+∈u(I);
where y' is the adjusted sample label, ε is the smoothing factor, y is the sample label before adjustment, and u (I) is the uniform distribution of the compliance class number I.
Optionally, the preprocessing the multi-beam data of the original sea area to obtain original back scattering data and original landform class label data includes:
decoding and correcting the multi-beam data of the original sea area to obtain temporary back scattering data and the original landform type tag data;
and denoising the temporary back-scattered data through a wavelet transformation function to obtain the original back-scattered data.
Compared with the prior art, the method for constructing the intelligent classification model of the submarine topography has the advantages that: firstly, preprocessing multi-beam data of an original sea area to obtain original back scattering data and original landform type tag data, training and optimizing an original classification model through the original back scattering data and the original landform type tag data, namely, constructing a model through an improved tag smooth cross entropy loss function and a pruning type dense connection network to obtain a seabed landform classification model, wherein the pruning type dense connection network is structurally characterized in that each layer uses input from all previous layers and transmits corresponding feature mapping of the pruning type dense connection network to all subsequent layers, compared with a traditional convolutional neural network, more global high-level features can be extracted by utilizing the pruning type dense connection network structure, meanwhile, the calculated amount is reduced, training can be performed more accurately and efficiently, in the model training process, the influence on the representation of the seabed landform features due to the similarity of landform types (such as a water channel and a depression) can be reduced based on the improved tag smooth cross entropy loss function, and the representation capability of the landform features is further improved; therefore, the application can effectively extract and fuse the space information of multiple dimensions through the combination of the pruning type dense connection network and the improved label smooth cross entropy loss function, not only can comprehensively extract the characteristics of the submarine landform, but also can improve the running efficiency of the model, and can also avoid the phenomenon that the final classification result is lost or inaccurate due to the similarity of a large number of landform categories, thereby improving the classification precision and reliability.
In order to solve the technical problems, the application also provides a device for constructing the intelligent classification model of the submarine topography, which comprises the following steps:
the acquisition unit is used for acquiring multi-beam data of the original sea area;
the processing unit is used for preprocessing the multi-beam data of the original sea area to obtain original back scattering data and original landform type label data;
the processing unit is also used for training and optimizing the original classification model according to the original back scattering data and the original landform class label data to obtain a submarine landform classification model; the original classification model is constructed based on an improved label smooth cross entropy loss function and a pruning type dense connection network, and the submarine topography classification model is used for predicting submarine topography categories of sea areas.
The device for constructing the intelligent classification model of the submarine topography has the same advantages as the method for constructing the intelligent classification model of the submarine topography compared with the prior art, and is not described in detail herein.
In order to solve the technical problems, the application also provides an intelligent classification method for the submarine topography, which comprises the following steps:
acquiring multi-beam data of a target sea area;
preprocessing the multi-beam data of the target sea area to obtain target back-scattering data;
inputting the target backscatter data into a submarine topography classification model obtained by the submarine topography intelligent classification model construction method to obtain a final submarine topography classification result.
The intelligent classification method for the submarine topography has the same advantages as the intelligent classification model construction method for the submarine topography compared with the prior art, and is not described in detail herein.
In order to solve the technical problems, the application also provides an intelligent classification device for submarine topography, which comprises:
the acquisition module is used for acquiring multi-beam data of the target sea area;
the processing module is used for preprocessing the multi-beam data of the target sea area to obtain target back scattering data;
the processing module is also used for inputting the target back scattering data into the submarine topography classification model obtained by the submarine topography intelligent classification model construction method to obtain a final submarine topography classification result.
The intelligent classification method for the submarine topography has the same advantages as the intelligent classification model construction method for the submarine topography compared with the prior art, and is not described in detail herein.
In order to solve the technical problems, the application also provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and when the computer program is read and run by a processor, the method for constructing the intelligent classification model of the submarine topography is realized, or the method for intelligently classifying the submarine topography is realized.
The advantages of the computer readable storage medium and the method for constructing the intelligent classification model of the submarine topography are the same as those of the prior art, and are not described in detail herein.
Detailed Description
In order that the above objects, features and advantages of the application will be readily understood, a more particular description of the application will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings.
As shown in fig. 1, in one embodiment, a method for constructing an intelligent classification model of a submarine topography is provided, which includes the following steps:
step S1, acquiring multi-beam data of an original sea area;
specifically, an original sea area is selected as a research area, multi-beam data with a submarine topography type label is collected, wherein the multi-beam data is obtained by simultaneously transmitting a plurality of transmitting frequencies (multi-frequency beams) and receiving scattered signals by a receiver. For example: after the transducer transmitting array of the multi-beam system excites acoustic energy downwards along the two sides of the ship, sound waves are propagated in sea water, when the sound waves meet a submarine interface, the sound waves are transmitted and scattered and returned to the transducer receiving array, and the transducer receiving array receives arrival angles of the sound waves, data during traveling and the like in real time, so that multi-beam data are obtained. Because sea water is an inhomogeneous acoustic medium, sound waves can change in advancing direction along with the direction anisotropy of the medium, and therefore, the inversion of the real sea bottom according to the arrival angle of the sound waves and the data and medium inhomogeneities during 'traveling' is the basic working principle of a multi-beam system. It should be noted that, when selecting the research area, a region with relatively comprehensive geomorphic categories covered by the submarine topography dataset may be selected, which is more beneficial to the subsequent model training process, for example, the Moro bay area and the wave male angle area of California may be selected as the research area.
Step S2, preprocessing the multi-beam data of the original sea area to obtain original back scattering data and original landform class label data;
specifically, decoding multi-beam data of an original sea area, performing geographic correction, radiation correction and the like to generate back scattering data and landform type tag data, wherein the landform type tag comprises a rock outcrop, a water channel, a depression, a ridge and a land frame, and the land frame is a gentle part of a continent extending into the sea, namely the continent covered by sea water; rock outcrop is the area of the bedrock exposed to the ocean surface; the depression is a general term for any relative concave part of the seabed, in particular a low-lying area surrounded by a height; waterways are linear or curved waterways over other flatter areas; the ridge is a long and narrow elevation, typically pointed and steep sided, with larger ridges forming extended elevations between valleys.
Step S3, training and optimizing an original classification model according to the original back scattering data and the original landform class label data to obtain a submarine landform classification model; the original classification model is constructed based on an improved label smooth cross entropy loss function and a pruning type dense connection network, and the submarine topography classification model is used for predicting submarine topography categories of sea areas.
Specifically, the original classification model is obtained by integrating an improved label smoothing cross entropy loss function into a pruning type dense connection network, a dataset is formed according to the original backscatter data and the original landform class label data, landform classes are divided according to Coastal and Marine Ecological Classification Standards (CMECS), a training set, a test set and a verification set are obtained, the original classification model is trained and optimized through the training set, the generalization performance of a trained final model (a submarine landform classification model) is checked through the test set, the performance of the final model is finally checked through the verification set, the improved label smoothing cross entropy loss function is a regularization method, and a weighted average is applied between uniform distribution and hard labels to generate soft labels so as to extract depth features, and similarity among classes is avoided.
According to the method for constructing the intelligent submarine topography classification model, original backscattering data and original topography type label data are obtained by preprocessing multi-beam data of an original sea area, the original classification model is trained and optimized through the original backscattering data and the original topography type label data, namely, the model constructed through an improved label smooth cross entropy loss function and a pruning type dense connection network is obtained, the submarine topography classification model is obtained, the structure characteristics of the pruning type dense connection network are that each layer uses input from all previous layers and transmits corresponding characteristic mapping of the pruning type dense connection network to all subsequent layers, compared with a traditional convolutional neural network, more global high-level characteristics can be extracted through the pruning type dense connection network structure, meanwhile, the calculated amount is reduced, training can be performed accurately and efficiently, in the model training process, influence on submarine topography characteristic representation due to topography type similarity (such as: submarine waterways and depression) can be reduced, and the characteristic representation capacity of the topography characteristic can be improved; therefore, the application can effectively extract and fuse the space information of multiple dimensions through the combination of the pruning type dense connection network and the improved label smooth cross entropy loss function, not only can comprehensively extract the characteristics of the submarine landform, but also can improve the running efficiency of the model, and can also avoid the phenomenon that the final classification result is lost or inaccurate due to the similarity of a large number of landform categories, thereby improving the classification precision and reliability.
In some embodiments, the process of constructing the original classification model includes:
step T1, obtaining an original dense connection network;
step T2, pruning operation is carried out on the original dense connection network, and the pruning dense connection network is obtained based on a downsampling function;
and step T3, obtaining the original classification model according to the improved label smooth cross entropy loss function and the pruning type dense connection network.
Specifically, since the dense connection of the original dense connection network DenseNet is redundant, the greatest influence of the dense connection is to influence the network efficiency, in order to reduce the redundancy problem of the DenseNet, the lightweight neural network CondenseNet is obtained by pruning the main network of the DenseNet, and the weight is pruned directly at the beginning of training instead of pruning the trained model, so that the network weight can be pruned accurately, training can be continued, and the network weight is smoother; compared with the dense connection of DenseNet which is only carried out in corresponding modules, the CondenseNet also connects the features among different modules, connection (jointing) operation is carried out, and for the problem of different sizes of feature graphs, down sampling operation is carried out based on a down sampling function through pooling operation, 1 x 1Conv (convolution) is replaced by 1 x 1L-Conv (learned group convolution), 3 x 3Conv is replaced by 3 x 3G-Conv (group convolution), so that the CondenseNet not only can extract the main topography features of the seabed, but also improves the operation efficiency, in addition, the Growth Rate (Growth Rate) is exponentially increased, the effect of high-level features on the model is larger for deep networks, and the network performance is improved by increasing the channel number of a rear layer; wherein learned group convolution is a learning group convolution, and the utilization of redundant features is greatly reduced by automatically learning the input feature group.
Specifically, in the structural hierarchy of the original classification model, each layer uses inputs from all previous layers and passes its corresponding feature map to all subsequent layers. These short connections between layers near the input and output allow the efficient passing of previous features to the back for automatic multiplexing of features, i.e. all previous layers are considered as input and then passed into the next layer, unlike some conventional network structures where there is a (l+1)/2 connection instead of a unidirectional connection in the L-layer convolutional layer, the convolutional layer operation formula in the dense block is as follows:
x L =T L ([x 0 ,x 1 ,…,x l-1 ]);
wherein x is 0 ,x 1 ,…,x l-1 Is the convolution layer of the first l layers, x L Is the convolution layer output, T L Is a set containing nonlinear transformations, including convolution, pooling, reLu layers (activation function layers).
Each dense block includes multiple sets of 1 x 1 and 3 x 3 convolutional layers with the same padding for concatenated operation. While this architecture uses a densely connected pattern, it requires fewer parameters than a conventional convolutional network. In fact, this network architecture eliminates the need to learn redundant information, reducing the number of natural resources required by the network layer. Thus, the parameter efficiency is significantly improved. On the other hand, the continuous connection of the different layers requires that each layer obtain gradients from the original input data and the loss function. This fast access improves the information flow between layers and reduces the gradient vanishing problem. The feature multiplexing method is beneficial to constructing a deeper network architecture and extracting deep semantic relations of feature interconnection.
In some embodiments, in step S3, training and optimizing the original classification model according to the original backscatter data and the original landform class label data to obtain a subsea landform classification model includes:
step S31, inputting the original back scattering data into the original classification model for training to obtain a temporary prediction classification result;
and step S32, performing tuning operation according to the temporary prediction classification result and the original landform class label data, and taking the tuned original classification model as the submarine landform classification model.
In some embodiments, in step S32, the performing an tuning operation according to the temporary prediction classification result and the original landform class label data, and taking the tuned original classification model as the submarine landform classification model includes:
step S321, carrying out loss calculation through the improved label smooth cross entropy loss function according to the temporary prediction classification result and the original landform class label data to obtain loss function output;
step S322, model parameters of the original classification model are adjusted according to the loss function output until the loss function input meets preset conditions, and the original classification model after parameter adjustment is used as the sea bottom landform classification model; wherein the improved label smoothing cross entropy loss function is obtained by a label smoothing function and a cross entropy loss function.
Specifically, depth feature extraction is performed on original backscatter data through an original classification model, a temporary prediction classification result is obtained, loss calculation is performed on the temporary prediction classification result and a real label (original landform class label data) through an improved label smooth cross entropy loss function, parameter adjustment is performed on the original classification model through obtained loss output until accuracy requirements are met, the adjusted original classification model is used as a final sea-bottom landform classification model, the improved label smooth cross entropy loss function is obtained by improving the cross entropy loss function, a label smooth concept is introduced, and the generated label smooth cross entropy loss function can avoid the problem of misclassification caused by similarity between classes in the training process. When the main features in different types of topography are the same or very similar, the similarity between the classes of the submarine topography dataset is displayed, and the model training is negatively influenced. In order to reduce the influence of the similarity between the categories, the traditional cross entropy loss function can be combined with label smoothing, so that the phenomenon that the final classification result is lost or inaccurate due to a large number of landform category similarities is avoided, and the classification precision and reliability are further improved.
In some embodiments, the improved label smoothing cross entropy loss function is obtained by a label smoothing function and a cross entropy loss function, comprising:
obtaining the improved label smoothing cross entropy loss function by carrying out fusion operation on the label smoothing function and the cross entropy loss function; wherein the tag smoothing function is:
y′=(1-∈)y+∈u(I);
where y' is the adjusted sample label, ε is the smoothing factor, y is the sample label before adjustment, and u (I) is the uniform distribution of the compliance class number I.
Specifically, for classification of images, a softmax function is typically added at the last layer to calculate the probability of each class of input predicted data, and a cross entropy loss function is used to calculate the loss value. And a category vector is typically converted to a one-hot vector (one-hot indicates that the subscript position of the word is set to 1 and the other positions are set to 0), where for an array of length n, only one element is 1 and the remaining elements are 0. This feature allows the generation of accurate and zero probabilities to encourage the gap between the true class and the other classes to be as large as possible, which means that the network model rewards the input image features of the correct labels and penalizes the wrong input image features. However, the feature gap between similar landform categories is relatively small, and this characteristic can lead to overfitting of the identification of features. Therefore, an improved label smooth cross entropy loss function is introduced, the influence of similar landform categories on landform feature representation is reduced, and the representation capability of the landform features is improved.
The traditional softmax formula is as follows:
;
wherein p is i Is the probability assigned to class i, w i Representing the weight and bias of the last layer, x is a vector containing depth features extracted from the image.
Re-use of back-propagation algorithm to calculate and minimize the actual target y i And network output p i The expected value of cross entropy between them is as follows:
;
wherein y is i A "1" indicates correct classification, and a "0" indicates incorrect classification.
It can be found that the loss function without tag smoothing only calculates the value of the correct position of the tag and does not calculate the wrong tag. This can result in the network being overly focused on increasing the probability of correct tags, rather than decreasing the probability of predicting false tags. The end result is that the model fits well to its own training set, but has poor results for other test sets. In particular, when a large number of seafloor landforms are similar and the loss of similar class labels is not considered, an overfitting situation is more likely to occur.
To take into account the loss values of the correct label position (i.e. the position of the one-hot label being "1") and the other wrong label positions (i.e. the position of the one-hot label being "0") in the training data samples, a label smoothing function is introduced, which is expressed as follows:
y′=(1-∈)y+∈u(I);
where y' is the adjusted sample label, ε is the smoothing factor, y is the sample label before adjustment, and u (I) is the uniform distribution of the compliance class number I.
In some embodiments, in step S2, the preprocessing the multi-beam data of the original sea area to obtain original backscatter data and original landform class label data includes:
s21, decoding and correcting the multi-beam data of the original sea area to obtain temporary back scattering data and the original landform type tag data;
and S22, denoising the temporary back-scattered data through a wavelet transformation function to obtain the original back-scattered data.
In some preferred embodiments, firstly, identifying and correcting the same area of multi-beam data of an original sea area according to tag data to obtain temporary back-scattering data and original landform type tag data, then converting the back-scattering data into tag data of TIFF, cutting the tag data into 64 x 64 data, deleting blank areas contained in the cut data, and finally obtaining original back-scattering data in TIFF format; the original back-scattered data and the original relief type tag data in the TIFF format are obtained by segmenting multi-beam data of an original sea area by combining Arcgis and python in a batch mode, and the multi-beam data and the remote sensing data are different, and the original back-scattered data and the original relief type tag data are back-scattered data which are converted into the TIFF format after a series of processes such as decoding, geographic correction and radiation correction are carried out on the multi-beam original data of the original sea area.
In some embodiments, as shown in fig. 2, acquiring multi-beam data of an original sea area, and performing operations such as decoding, geographic correction, radiation correction and the like on the multi-beam data of the original sea area to obtain original backscatter data and original landform class label data; the method comprises the steps of inputting original backscatter data into an original classification model for training, namely performing 2-dimensional convolution, maximum pooling and 3-time iterative dimension reduction (an iterative module is used for performing dimension reduction operation, wherein the iterative module is formed by performing downsampling operation through a CondenseNet and based on a downsampling function) and global average pooling, obtaining a temporary prediction classification result, performing loss calculation through an improved label smooth cross entropy loss function according to the temporary prediction classification result and original landform class label data, and performing parameter tuning on the original classification model through back propagation, wherein the parameter tuning is equivalent to the CondenseNet, and finally obtaining a submarine landform classification model; the method has the advantages that the CondenseNet is obtained by pruning the backbone network of the DenseNet, and the CondenseNet is combined with the downsampling operation for multiple iterations, so that the calculated amount is further reduced while the calculated amount is reduced on the basis of the CondenseNet, the operation efficiency is further improved, and the number of iteration modules, namely the number of iterations, is not limited and can be determined according to actual conditions.
According to the method for constructing the intelligent submarine topography classification model, original backscattering data and original topography type label data are obtained by preprocessing multi-beam data of an original sea area, the original classification model is trained and optimized through the original backscattering data and the original topography type label data, namely, the model constructed through an improved label smooth cross entropy loss function and a pruning type dense connection network is obtained, the submarine topography classification model is obtained, the structure characteristics of the pruning type dense connection network are that each layer uses input from all previous layers and transmits corresponding characteristic mapping of the pruning type dense connection network to all subsequent layers, compared with a traditional convolutional neural network, more global high-level characteristics can be extracted through the pruning type dense connection network structure, meanwhile, the calculated amount is reduced, training can be performed accurately and efficiently, in the model training process, influence on submarine topography characteristic representation due to topography type similarity (such as: submarine waterways and depression) can be reduced, and the characteristic representation capacity of the topography characteristic can be improved; therefore, the application can effectively extract and fuse the space information of multiple dimensions through the combination of the pruning type dense connection network and the improved label smooth cross entropy loss function, not only can comprehensively extract the characteristics of the submarine landform, but also can improve the running efficiency of the model, and can also avoid the phenomenon that the final classification result is lost or inaccurate due to the similarity of a large number of landform categories, thereby improving the classification precision and reliability.
As shown in fig. 3, another embodiment of the present application provides a device for constructing an intelligent classification model of a submarine topography, including:
the acquisition unit is used for acquiring multi-beam data of the original sea area;
the processing unit is used for preprocessing the multi-beam data of the original sea area to obtain original back scattering data and original landform type label data;
the processing unit is also used for training and optimizing the original classification model according to the original back scattering data and the original landform class label data to obtain a submarine landform classification model; the original classification model is constructed based on an improved label smooth cross entropy loss function and a pruning type dense connection network, and the submarine topography classification model is used for predicting submarine topography categories of sea areas.
Yet another embodiment of the present application provides a submarine topography classification model construction apparatus, including a memory and a processor: a memory for storing a computer program; and the processor is used for realizing the method for constructing the intelligent classification model of the submarine topography when executing the computer program.
It should be noted that the device may be a computer device such as a server, a mobile terminal, or the like.
As shown in fig. 4, still another embodiment of the present application further provides an intelligent classification method for a submarine topography, including the following steps:
step A1, acquiring multi-beam data of a target sea area;
step A2, preprocessing the multi-beam data of the target sea area to obtain target back scattering data;
and step A3, inputting the target back scattering data into a submarine topography classification model obtained by the submarine topography intelligent classification model construction method to obtain a final submarine topography classification result.
Specifically, acquiring multi-beam data of a target sea area, namely multi-beam data of the sea area to be detected, obtaining target back-scattering data by decoding, geographic correction, radiation correction and other operations on the multi-beam data of the sea area to be detected, and inputting the target back-scattering data into a submarine topography classification model obtained by the submarine topography intelligent classification model construction method to obtain a submarine topography classification result of the sea area to be detected.
In this embodiment, multi-beam data of a research sea area (the landform class of the research sea area is 5 landform classes including rock outcrop, water channel, land frame, depression and ridge) are selected, multi-beam data of the research sea area are decoded, geographically corrected, radiation corrected and the like to obtain back scattering data, the back scattering data are divided to obtain a training set, a test set and a verification set, an original classification model is iteratively trained on the training set for 20 times, an average value is obtained, positive and negative deviations are calculated to obtain a submarine landform classification model, accuracy evaluation is carried out on the submarine landform classification model through the test set to obtain accuracy of about 71.74 +/-0.69%, recall rate is 50.62 +/-0.72%, accuracy 70.81 +/-0.84% higher than that based on Densenet, recall rate is 45.52+/-0.63% and accuracy 69.03 +/-0.58% higher than that based on Resnet, recall rate 42.83 +/-0.67% is obtained through comparison, and the similarity of landform classification is proved to be greatly improved.
Still another embodiment of the present application provides an intelligent classification apparatus for seafloor topography, including:
the acquisition module is used for acquiring multi-beam data of the target sea area;
the processing module is used for preprocessing the multi-beam data of the target sea area to obtain target back scattering data;
the processing module is also used for inputting the target back scattering data into the submarine topography classification model obtained by the submarine topography intelligent classification model construction method to obtain a final submarine topography classification result.
It should be noted that the device may be a computer device such as a server, a mobile terminal, or the like.
FIG. 5 illustrates an internal block diagram of a computer device in one embodiment. The computer device includes a processor, a memory, a network interface, an input device, and a display screen connected by a system bus. The memory includes a nonvolatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system, and may also store a computer program that, when executed by a processor, causes the processor to implement a subsea topography intelligent classification method. The internal memory may also store a computer program which, when executed by the processor, causes the processor to execute the method for intelligently classifying the seafloor topography. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, the input device of the computer equipment can be a touch layer covered on the display screen, can also be keys, a track ball or a touch pad arranged on the shell of the computer equipment, and can also be an external keyboard, a touch pad or a mouse and the like.
In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, which when executed by a processor implements the method for constructing a smart classification model of a seafloor topography, or implements the method for smart classification of a seafloor topography.
Those skilled in the art will appreciate that all or part of the processes in the methods of the above embodiments may be implemented by a computer program for instructing relevant hardware, where the program may be stored in a non-volatile computer readable storage medium, and where the program, when executed, may include processes in the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.
It should be noted that in this document, terms such as "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The foregoing is only a specific embodiment of the application to enable those skilled in the art to understand or practice the application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Although the application is disclosed above, the scope of the application is not limited thereto. Various changes and modifications may be made by one skilled in the art without departing from the spirit and scope of the application, and these changes and modifications will fall within the scope of the application.