CN116738324B

CN116738324B - Model training method and identification method for single-towing operation behavior of fishing boat

Info

Publication number: CN116738324B
Application number: CN202311007860.3A
Authority: CN
Inventors: 毛伟
Original assignee: Taiji Computer Corp Ltd
Current assignee: Taiji Computer Corp Ltd
Priority date: 2023-08-11
Filing date: 2023-08-11
Publication date: 2023-12-22
Anticipated expiration: 2043-08-11
Also published as: CN116738324A

Abstract

The present disclosure proposes a model training method for identifying a single-drag operation behavior of a fishing boat and an identification method of a single-drag operation behavior of a fishing boat, the method comprising obtaining initial historical track data of a sample fishing boat having the single-drag operation behavior; filtering the initial historical track data to obtain target historical track data of the sample fishing boat in single-towing operation; the target historical track data are data of the sample fishing boat in single-towing operation; training the recognition model according to the target historical track data to obtain a trained target recognition model; the recognition model is used for recognizing whether the single drag operation acts or not based on the target historical track data. Therefore, training of the identification model can be achieved, and the prediction effect of the model is improved, namely the accuracy and reliability of the prediction result of the model are improved.

Description

Model training method and identification method for single-towing operation behavior of fishing boat

Technical Field

The disclosure relates to the technical field of computers, in particular to a model training method for identifying single-towing operation behaviors of a fishing boat and an identification method for the single-towing operation behaviors of the fishing boat.

Background

Currently, when a fishing boat is used for offshore fishing, a fishing license needs to be published in advance and the type of the fishing operation needs to be registered. However, during actual offshore operations, there may be situations where the rules are not complied with by some fishing vessels, such as when the operation mode is an unauthorized change to a single trawl operation mode. It should be noted that, the illegal operation may adversely affect the fishery resources and the marine ecology environment. Therefore, it is necessary to identify the operation of the fishing vessel to assist the relevant personnel in supervising the fishing vessel at sea.

Disclosure of Invention

The present disclosure proposes a model training method for identifying a fishing boat single drag operation behavior and a fishing boat single drag operation behavior identification method.

An embodiment of a first aspect of the present disclosure proposes a model training method for identifying a single drag operation behavior of a fishing vessel, the method comprising: acquiring initial historical track data of a sample fishing boat with single-towing operation behaviors; filtering the initial historical track data to obtain target historical track data; the target historical track data are data of the sample fishing boat in single-towing operation; training the recognition model according to the target historical track data to obtain a trained target recognition model; the recognition model is used for recognizing whether the target historical track data has the single-drag operation behavior or not.

An embodiment of a second aspect of the present disclosure provides a method for identifying a single-towing operation behavior of a fishing boat, where the method includes: acquiring operation behavior data of a target fishing boat; the operation behavior data is identified by using the target identification model obtained by the method according to the embodiment of the first aspect, so as to determine whether the target fishing boat has a single-towing operation behavior.

An embodiment of a third aspect of the present disclosure proposes a model training apparatus for identifying a single drag operation behavior of a fishing vessel, the apparatus comprising:

the acquisition module is used for acquiring initial historical track data of the sample fishing boat with the single-towing operation behavior;

the filtering module is used for filtering the initial historical track data to obtain target historical track data; the target historical track data are data of the sample fishing boat in single-towing operation;

the training module is used for training the recognition model according to the target historical track data so as to obtain a trained target recognition model; the recognition model is used for recognizing whether the target historical track data has the single-drag operation behavior or not.

An embodiment of a fourth aspect of the present disclosure provides a device for identifying a single-towing operation behavior of a fishing boat, where the device includes:

The acquisition module is used for acquiring the operation behavior data of the target fishing boat;

the identifying module is configured to identify the operational behavior data by using a target identifying model obtained by a method according to an embodiment of the first aspect, so as to determine whether the target fishing boat has a single-towing operational behavior.

An embodiment of a fifth aspect of the present disclosure proposes an electronic device, including: a memory, a processor; the memory has stored therein computer instructions which, when executed by the processor, implement the methods disclosed by embodiments of the first aspect of the present disclosure, or the methods disclosed by embodiments of the second aspect of the present disclosure.

An embodiment of a sixth aspect of the present disclosure proposes a computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when executed by a processor, implements the method disclosed by the embodiment of the first aspect of the present disclosure, or the method disclosed by the embodiment of the second aspect of the present disclosure.

According to the model training method for identifying the single-towing operation behavior of the fishing boat, initial historical track data of a sample fishing boat with the single-towing operation behavior are obtained; filtering the initial historical track data to obtain target historical track data; the target historical track data are data of the sample fishing boat in single-towing operation; training the recognition model according to the target historical track data to obtain a trained target recognition model; the recognition model is used for recognizing whether the single drag operation acts or not based on the target historical track data. Therefore, training of the identification model can be achieved, and the prediction effect of the model is improved, namely the accuracy and reliability of the prediction result of the model are improved.

Other effects of the above alternative will be described below in connection with specific embodiments.

Drawings

The drawings are for a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

fig. 1 is a flowchart of a model training method for identifying a single-towing operation of a fishing vessel according to an embodiment of the present disclosure.

Fig. 2 is a flowchart of a model training method for identifying a single-towing operation of a fishing vessel according to a second embodiment of the present disclosure.

Fig. 3 is a schematic diagram of an internal structure of the LSTM provided in the present disclosure.

Fig. 4 is a schematic diagram of a model structure of Bi-LSTM provided by the present disclosure.

FIG. 5 is a schematic diagram of an Attention mechanism model provided by the present disclosure.

Fig. 6 is a flowchart of a model training method for identifying a single-towing operation behavior of a fishing vessel according to a third embodiment of the present disclosure.

Fig. 7 is a schematic structural diagram of a spatio-temporal dual-flow fusion network provided by the present disclosure.

Fig. 8 is a flowchart of a model training method for identifying a single-towing operation behavior of a fishing boat according to a fourth embodiment of the present disclosure.

Fig. 9 is a flowchart of a method for identifying a single-towing operation behavior of a fishing boat according to a fifth embodiment of the present disclosure.

Fig. 10 is a schematic structural diagram of a model training device for identifying a single-towing operation of a fishing vessel according to a sixth embodiment of the present disclosure.

Fig. 11 is a schematic structural diagram of a device for identifying a single-towing operation behavior of a fishing boat according to a seventh embodiment of the present disclosure.

Fig. 12 is a block diagram of an electronic device according to one embodiment of the present disclosure.

Detailed Description

Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and intended for the purpose of explaining the present disclosure and are not to be construed as limiting the present disclosure.

The method and the device have the advantages that the data acquisition, storage, use, processing and the like in the technical scheme meet the relevant regulations of national laws and regulations.

Currently, the relevant departments coordinate marine traffic and supervise marine activities by employing a new navigation aid system, a ship automatic identification system (Automatic Identification System, abbreviated as AIS). For example, in a marine ranch fishing forbidden period, AIS equipment may be used to analyze whether there is illegal fishing activity based on fishing vessel trajectory data. The current AIS equipment can acquire the fishing boat track information such as the position, the speed and the ship number of the ship in near real time, but the system lacks analysis on the fishing boat track data, so that related staff cannot evidence illegal fishing behaviors of offshore ocean to implement effective supervision. With the explosive growth of data, the use of traditional manual inspection patterns to identify illegal fishing activities is inefficient and presents significant challenges to the supervision of the relevant authorities. Therefore, how to automatically analyze valuable information such as behavior and state change rules of the fishing boat from massive fishing boat track data is important to provide assistance and basis for relevant departments to process and decide.

At present, the fishing boat behavior recognition method in the related art mainly has the following three problems:

1. the manual extraction of corresponding manual features from the relevant data is very dependent on expert experience, and some important features are very easy to lose.

2. Multiple angle analysis of fishing vessel trajectory data is lacking. In some fishing boat behavior recognition methods, only fishing boat track data is regarded as time sequence data, and the characteristics of the geometric shape of the boat track are easily ignored by the methods; in other fishing boat behavior recognition methods, only the fishing boat trajectory data is regarded as image data, and the time sequence characteristics of dynamic attribute changes in the boat motion process are easily ignored by the methods.

In view of at least one of the above problems, the present disclosure proposes a model training method for identifying a single drag operation behavior of a fishing vessel and an identification method of a single drag operation behavior of a fishing vessel.

A model training method for identifying a fishing boat single drag operation behavior and an identification method of a fishing boat single drag operation behavior according to embodiments of the present disclosure are described below with reference to the accompanying drawings.

It should be noted that, the execution body of the model training method for identifying the single-towing operation behavior of the fishing boat provided by the embodiment of the present disclosure is a model training device for identifying the single-towing operation behavior of the fishing boat, and the model training device for identifying the single-towing operation behavior of the fishing boat may be implemented by software and/or hardware.

The electronic device in the embodiment of the present disclosure may include a terminal device, a server, and the like, for example, the terminal device may be a hardware device having various operating systems, touch screens, and/or display screens, such as a mobile phone, a tablet computer, and a personal digital assistant, which is not specifically limited in this embodiment. In this example, an electronic device is taken as a server for example.

As shown in fig. 1, the model training method for identifying the behavior of a fishing boat single towing operation may include:

step 101, obtaining initial historical track data of a sample fishing boat with single-towing operation behaviors.

In embodiments of the present disclosure, initial historical trajectory data for a sample fishing vessel with single-towing operation behavior may be obtained. For example, initial historical trajectory data of a sample fishing vessel with single towing operation behavior can be obtained through radar, satellite navigation systems, AIS equipment and the like.

It should be noted that the above example of acquiring the initial historical track data is merely exemplary, and in practical application, the initial historical track data may also be acquired by other manners, which is not limited in this disclosure.

It should be further noted that the number of sample fishing boats is not limited in the present disclosure, and the number of sample fishing boats may be one or may be plural.

In one possible implementation manner of the embodiment of the present disclosure, after the initial historical track data is acquired, a list of operable areas may be acquired, and the initial historical track data is screened according to the operable areas in the list of operable areas, so as to preserve the initial historical track data of the sample fishing boat whose activity range is not in the operable areas.

In embodiments of the present disclosure, the workable area may be used to indicate an area where a fishing vessel may perform a single towing operation.

In the embodiment of the present disclosure, one operable area may be included in the operable area list, or a plurality of operable areas may be included, which is not limited by the present disclosure.

It is understood that the list of workable areas may be obtained, for example, the list of workable areas may be stored in advance, so that the executive body of the present disclosure may read the list of workable areas.

In the embodiment of the disclosure, the initial historical track data can be screened according to the operable area in the operable area list so as to keep the initial historical track data of the sample fishing boat with the activity range not in the operable area.

As an example, first, position data corresponding to an operable area in an operable area list may be acquired; secondly, the initial historical track data of the sample fishing boat can be compared with the position data corresponding to the operable area, so that the initial historical track data of the sample fishing boat which does not belong to the operable area can be screened out from the initial historical track data; finally, the initial historical track data of the sample fishing boat which does not belong to the operable area can be reserved.

In another possible implementation of the embodiments of the present disclosure, the activity range of the sample fishing vessel may be screened according to the initial historical track data to preserve the initial historical track data of the sample fishing vessel with the activity range within a set distance range from the shore and/or port.

It should be noted that the set distance may be preset, for example, may be 15km, 20km, or the like, which is not limited in the present disclosure.

In yet another possible implementation of the disclosed embodiments, the initial historical track data may be pre-processed. It should be noted that, the preprocessing may include data cleansing, data integration, data transformation, data protocol, and the like, which is not limited in this disclosure.

Step 102, filtering the initial historical track data to obtain target historical track data.

The target historical track data can be data of the sample fishing boat in single-towing operation.

It will be appreciated that noise data may be present in the initial historical track data obtained. Thus, in the present disclosure, the initial historical track data may be filtered to obtain the target historical track data.

And step 103, training the recognition model according to the target historical track data to obtain a trained target recognition model.

It should be noted that, the recognition model may be used to recognize whether there is a single-drag operation based on the target historical track data.

In the embodiment of the disclosure, the recognition model can be trained according to the target historical track data of the sample fishing boat so as to obtain a trained target recognition model.

In order to clearly illustrate how the above embodiments of the present disclosure train the recognition model according to the target historical track data to obtain a trained target recognition model, the present disclosure also provides a model training method for recognizing the single towing operation behavior of a fishing boat.

As shown in fig. 2, the model training method for identifying the behavior of a fishing boat single towing operation may include:

step 201, obtaining initial historical track data of a sample fishing boat with single-towing operation behaviors.

Step 202, filtering the initial historical track data to obtain target historical track data.

The execution of steps 201 to 202 may refer to the execution of any embodiment of the present disclosure, and will not be described herein.

Step 203, inputting the target history track data into a first network of the recognition model to obtain the space-time fusion feature.

Wherein the first network may be used to extract spatiotemporal fusion features.

In an embodiment of the disclosure, the recognition model may have a first network, and the target historical track data may be input into the first network of the recognition model, so that a spatio-temporal fusion feature may be obtained.

Step 204, inputting the space-time fusion feature into a second network of the recognition model to obtain a time sequence feature.

In embodiments of the present disclosure, the recognition model may have a second network, and the spatio-temporal fusion feature may be input into the second network of the recognition model, so that the timing feature may be obtained.

In one possible implementation of the disclosed embodiments, the second network may be a Bi-LSTM (Bidirectional Long Short-Term Memory Network, bi-directional long short-term memory network).

As an example, a second network with Bi-LSTM as the recognition model is illustrated. It should be noted that Bi-LSTM is composed of a forward propagating LSTM (Long Short-Term Memory) and a backward propagating LSTM (Long Short-Term Memory network), wherein the internal structure of the LSTM is shown in fig. 3.

As shown in fig. 3, the core of the LSTM is a memory cell that functions to select useful information and remove redundant information. Meanwhile, the LSTM may control the propagation capability between information and the memory unit through its special gate structure, where the gates of the LSTM include an input gate, a forget gate and an output gate, and the structure of any of the gates may use a Sigmoid function, for example, the Sigmoid function is represented by the following formula (1):

σ(x) =(1+e ^-x ) ^-1 ；（1）

the value range of the output value of the Sigmoid function is (0, 1), which can be used for reflecting the information storage condition.

The input/output process of the memory cell of the LSTM may be specifically expressed as:

i _t =σ(W _i ·[x _t ,H _t-1 ]+b _i ) ；（2）

f _t =σ(W _f ·[x _t ,H _t-1 ]+b _f ) ；（3）

o _t =σ(W _o ·[x _t ,H _t-1 ]+b _o ) ；（4）

g _t =tanh(W _c ·[x _t ,H _t-1 ] +b _c )；（5）

c _t = f _t * c _t-1 +i _c * g _t ；(6)

H _t = o _t * tanh(c _t )； (7)

wherein c _t The state of the memory unit at the current time t; c _t-1 The state of the memory unit at the previous time t-1; x is x _t Is the input value of the memory cell; h _t-1 An output value indicating the last time; w (W) _i B is a weight vector corresponding to the input gate _i Bias corresponding to the input gate; w (W) _f Weight vector corresponding to forget gate, b _f Bias corresponding to the forgetting door; w (W) _o To output the weight vector corresponding to the gate b _o Bias corresponding to the output gate; w (W) _c And b _c Is a parameter that can be learned; i.e _t Is the value of the input gate and can determine the information that remains; f (f) _t Is the value of the forgetting gate and can determine the deleted information; o (o) _t Is the value of the output gate and can determine the output of the memory cell; g _t Is a candidate value vector.

In equation (7), the output H of LSTM _t Whether the memory cell is activated or not is controlled by the output gate.

In the Bi-LSTM, the input of the current moment depends on the single drag behavior track frames of the fishing boat at the previous moment and the next moment, so that the time sequence information of the single drag behavior track frames of the fishing boat can be fully considered, and the model structure of the Bi-LSTM is shown in the figure 4.

In FIG. 4, reference numeral w _i (i=1, 2,., 6) is the weight of the corresponding cell layer; x is x _t Space-time fusion features extracted for a first network of the recognition model; h is a _t Representing LSTM cells { … …, x formed from front-to-back signature sequences _t-1 ，x _t ，x _t+1 ，……}；h ^' _t Represents LSTM cell { … …, x formed by the back-to-front signature sequence _{t +1} ，x _t ，x _{t -1} ，…… } ；

h _t =f(w ₁ x _t +w ₂ h _t+1 +b ₁ )；(8)

h ^' _t = f (w ₃ x _t +w ₅ h _t+1 +b ₂ )；(9)

O ^' _t =g(w ₄ h _t +b ₃ )；(10)

O ^" _t = g (w ₆ h ^' _t +b ₃ )；(11)

O _t = (O ^' _t +O ^" _t ) /2；(12)

Wherein, f (·) and g (·) are both activation functions; b ₁ And b ₃ Is biased; o's' _t And O' _t Processing the result of the space-time fusion characteristic for the LSTM unit at the corresponding moment; o (O) _t Is the timing characteristic of the Bi-LSTM output.

It should be noted that, compared with the conventional LSTM model, the Bi-LSTM algorithm can learn the track information of the single drag behavior of the fishing boat at the previous and subsequent moments at a certain moment at the same time, so that the timing characteristics or the timing information obtained by the Bi-LSTM is more robust.

Step 205, inputting the time series features into the attention network of the recognition model to obtain weighted fusion features.

In embodiments of the present disclosure, the recognition model may have an attention network, and the temporal features may be input into the attention network of the recognition model to obtain weighted fusion features.

In one possible implementation of the disclosed embodiments, the attention network may include a hidden layer and a fully connected layer.

In another possible implementation of the embodiments of the present disclosure, where the attention network includes a hidden layer and a fully connected layer, first, the timing characteristics may be input into the hidden layer of the attention network to obtain intermediate characteristics; finally, the intermediate features may be input into a fully connected layer of the attention network to weight the intermediate features to obtain weighted fusion features.

As an example, attention networks are exemplified for the Attention mechanism model. FIG. 5 is a schematic diagram of an Attention mechanism model. As shown in FIG. 5, O _t A t-th timing feature output for the second network of the identified model; intermediate features S are obtained after passing through a hidden layer in the Atention mechanism model _t The method comprises the steps of carrying out a first treatment on the surface of the There may be:

e _t =tanh(w _t s _t +b _t )；(13)

；(14)

；(15)

wherein w is _t B is a weight parameter _t Is biased; e, e _t Is an attention score; alpha _t Is a weight parameter; y is a weighted fusion feature.

And 206, identifying whether the sample fishing boat has single-towing operation behaviors or not by adopting an output layer of the identification model according to the weighted fusion characteristics.

In embodiments of the present disclosure, the recognition model may have an output layer, such as the output layer may employ a Softmax function.

In the embodiment of the disclosure, whether the sample fishing boat has the single-towing operation behavior can be identified by adopting an output layer of the identification model according to the weighted fusion characteristics. For example, the weighted fusion characteristics can be input into an output layer of the identification model, and whether the sample fishing boat has the single-towing operation behavior can be identified according to the output result of the output layer.

Step 207, performing parameter adjustment on the recognition model based on the difference between the recognition result and the actual result to obtain a trained target recognition model.

In embodiments of the present disclosure, the recognition result may indicate whether the sample fishing vessel has a single drag operation behavior.

In embodiments of the present disclosure, the actual results may indicate that the sample fishing vessel has single towing operation behavior.

In the embodiment of the disclosure, parameter adjustment can be performed on the recognition model based on the difference between the recognition result and the actual result to obtain a trained target recognition model.

As an example, the loss value may be generated according to a difference between the recognition result and the actual result, where the loss value and the difference are in a forward relationship (i.e., in a positive correlation relationship), that is, the smaller the difference is, the smaller the value of the loss value is, the larger the difference is, and the larger the value of the loss value is. Therefore, the parameter adjustment can be carried out on the identification model according to the loss value so as to minimize the value of the loss value.

It should be noted that, the foregoing example is only implemented by taking the termination condition of training of the recognition model as the minimizing of the loss value, and other termination conditions may be set in practical application, for example, the termination condition may be that the training number reaches the set number, or the termination condition may be that the training duration reaches the set duration, the loss value converges, and so on, which is not limited in the disclosure.

According to the model training method for identifying the single-towing operation behavior of the fishing boat, the target historical track data is input into a first network of an identification model to obtain space-time fusion characteristics; the first network is used for extracting space-time fusion characteristics; inputting the space-time fusion features into a second network of the recognition model to obtain time sequence features; inputting the time sequence characteristics into an attention network of the recognition model to obtain weighted fusion characteristics; according to the weighted fusion characteristics, an output layer of the identification model is adopted to identify whether the sample fishing boat has a single-towing operation behavior; and carrying out parameter adjustment on the recognition model based on the difference between the recognition result and the actual result so as to obtain a trained target recognition model. Therefore, based on the deep learning technology, the recognition model is trained, so that the recognition model can accurately learn the time-space fusion characteristics, the time sequence characteristics and the weighted fusion characteristics of the single-towing operation behaviors of the sample fishing boat based on the target historical track data, whether the sample fishing boat has the single-towing operation behaviors or not can be recognized based on the recognition result of the output layer of the recognition model, and the stability of the model and the accuracy and the reliability of the model prediction result can be improved.

In order to clearly illustrate how the target historical track data is input into the first network of the identification model in any of the above embodiments of the present disclosure to obtain the space-time fusion feature, the present disclosure further proposes a model training method for identifying the behavior of a fishing boat single towing operation.

As shown in fig. 6, the model training method for identifying the behavior of a fishing boat single drag operation may include:

step 601, obtaining initial historical track data of a sample fishing boat with single-towing operation behaviors.

Step 602, filtering the initial historical track data to obtain target historical track data.

The execution of steps 601 to 602 may refer to the execution of any embodiment of the present disclosure, and will not be described herein.

Step 603, inputting the target historical track data into a time flow network of the first network to obtain a time characteristic.

It should be noted that, in this embodiment, the explanation of the first network may refer to any embodiment of the disclosure, which is not described herein.

In the disclosed embodiments, the first network may have a time-streaming network.

As one example, the time-flow network may be a convolutional neural network.

In one possible implementation of an embodiment of the present disclosure, a time flow network may include a first feature extraction layer, a second feature extraction layer, a third feature extraction layer, a fourth feature extraction layer, and a fifth feature extraction layer in series.

In embodiments of the present disclosure, the target historical track data may be input into a time-flow network of the first network to obtain the time signature.

Step 604, based on the target historical track data, performing feature extraction by using the spatial stream network and the temporal stream network of the first network to obtain spatial features.

In the disclosed embodiments, the first network may have a spatial stream network.

As an example, the spatial stream network may be a convolutional neural network.

In one possible implementation of an embodiment of the disclosure, the spatial stream network may include a sixth feature extraction layer, a seventh feature extraction layer, an eighth feature extraction layer, a ninth feature extraction layer, a tenth feature extraction layer, a first fused feature layer, and a second fused feature layer.

In the embodiment of the disclosure, the spatial stream network and the time stream network of the first network may be adopted to perform feature extraction based on the target historical track data, so that spatial features may be obtained.

As one possible implementation manner, in the case that the time-stream network includes a first feature extraction layer, a second feature extraction layer, a third feature extraction layer, a fourth feature extraction layer, and a fifth feature extraction layer that are connected in series, the space-stream network includes a sixth feature extraction layer, a seventh feature extraction layer, an eighth feature extraction layer, a ninth feature extraction layer, a tenth feature extraction layer, a first fused feature layer, and a second fused feature layer that are connected in series, the target historical track data and the first feature may be input into the sixth feature extraction layer of the space-stream network to obtain a sixth feature; wherein the first feature may be an output of the first feature extraction layer; the sixth feature and the second feature may be input into a seventh feature extraction layer of the spatial stream network to obtain a seventh feature; wherein the second feature may be an output of the second feature extraction layer; the seventh feature and the third feature may be input into an eighth feature extraction layer of the spatial stream network to obtain an eighth feature; wherein the third feature may be an output of the third feature extraction layer; the eighth feature and the fourth feature may be input into a ninth feature extraction layer of the spatial stream network to obtain a ninth feature; wherein the fourth feature may be an output of the fourth feature extraction layer; the ninth feature and the fifth feature may be input into a tenth feature extraction layer of the spatial stream network to obtain a tenth feature; wherein the fifth feature may be an output of the fifth feature extraction layer; the tenth feature can be input into a first feature fusion layer of the spatial stream network to obtain a first fusion feature; the first fused feature may be input into a second feature fusion layer of the spatial stream network to obtain a spatial feature.

Step 605, fusing the temporal features and the spatial features to obtain space-time fusion features.

In the embodiment of the disclosure, the temporal features and the spatial features can be fused, and the space-time fusion features can be obtained.

As an example, the temporal features and the spatial features may be weighted and fused, resulting in a spatio-temporal fusion feature.

As another example, temporal features and spatial features may be stitched and spatio-temporal fusion features may be derived.

As an example, a spatio-temporal dual stream converged network is exemplified as the first network. Fig. 7 is a schematic structural diagram of a spatio-temporal dual-flow fusion network provided by the present disclosure. As shown in fig. 7, the spatio-temporal dual-flow fusion network includes a temporal convolutional neural network (labeled as a temporal flow network in the present disclosure) and a spatial convolutional neural network (labeled as a spatial flow network in the present disclosure), wherein the temporal convolutional neural network includes a first feature extraction layer, a second feature extraction layer, a third feature extraction layer, a fourth feature extraction layer, a fifth feature extraction layer, a third fusion feature layer, and a fourth fusion feature layer connected in series, and the spatial convolutional neural network includes a sixth feature extraction layer, a seventh feature extraction layer, an eighth feature extraction layer, a ninth feature extraction layer, a tenth feature extraction layer, a first fusion feature layer, and a second fusion feature layer connected in series.

The target historical track data is input into a time convolution neural network, so that a time characteristic can be obtained; inputting target historical track data into a first feature extraction layer of a time convolution neural network to obtain first features; inputting the first feature into a second feature extraction layer of the time convolution neural network to obtain a second feature; inputting the second characteristic into a third characteristic extraction layer of the time convolution neural network to obtain a third characteristic; inputting the third feature into a fourth feature extraction layer of the time convolution neural network to obtain a fourth feature; and inputting the fourth feature into a fifth feature extraction layer of the time convolution neural network to obtain a fifth feature.

Inputting the target historical track data and the first characteristic into a sixth characteristic extraction layer of the spatial convolution neural network, so as to obtain a sixth characteristic; inputting the sixth feature and the second feature into a seventh feature extraction layer of the spatial convolutional neural network, so as to obtain a seventh feature; inputting the seventh feature and the third feature into an eighth feature extraction layer of the spatial convolutional neural network, so as to obtain an eighth feature; inputting the eighth feature and the fourth feature into a ninth feature extraction layer of the spatial convolution neural network to obtain a ninth feature; inputting the ninth feature and the fifth feature into a tenth feature extraction layer of the spatial convolution neural network to obtain a tenth feature; inputting the tenth feature into a first feature fusion layer of the spatial convolution neural network to obtain a first fusion feature; and inputting the first fusion characteristic into a second characteristic fusion layer of the spatial convolution neural network to obtain the spatial characteristic.

After the time characteristic is obtained through the time convolution neural network of the space-time double-flow fusion network and the space characteristic is obtained through the space convolution neural network of the space-time double-flow fusion network, the time characteristic and the space characteristic can be fused to obtain the space-time fusion characteristic.

It can be understood that the space-time double-flow fusion network extracts apparent motion information (i.e., spatial features) through a spatial convolution neural network and extracts long-time motion information (i.e., temporal features) through a temporal convolution neural network, so that the space-time double-flow fusion network can identify the single towing behavior of the fishing vessel by utilizing the correlation of the spatial features and the temporal features. In the space-time double-flow fusion network, a unidirectional connection from time flow to space flow is adopted, the time characteristics extracted by the time flow are input into the space flow, and the time characteristics are associated with the space characteristics in the mode, so that the more deep single-towing behavior characteristics of the fishing boat can be extracted.

It should be noted that, although each layer inside the space-time double-flow fusion network is connected in one way, each layer also performs relevance learning, since the dynamic feature is important information of the single towing behavior of the fishing boat, the output feature of the time convolution neural network is fused with the output feature of the space convolution neural network, that is, the space-time fusion feature obtained by fusing the time feature and the space feature is used as the dominant feature of the recognition model of the disclosure to recognize the single towing behavior of the fishing boat.

In conclusion, when the space-time double-flow fusion network is adopted to extract the space-time fusion characteristics, a unidirectional connection mode is adopted between the time convolution neural network and the space convolution neural network in the recognition model, and the correlation learning of the long-time motion information and the apparent motion information is carried out, so that the characteristic expression capability of the recognition model can be improved, and the recognition accuracy of the recognition model can be further improved.

Step 606, the spatio-temporal fusion features are input into a second network of recognition models to obtain timing features.

In step 607, the temporal features are input into the attention network of the recognition model to obtain weighted fusion features.

Step 608, according to the weighted fusion characteristics, an output layer of the identification model is adopted to identify whether the sample fishing boat has single-towing operation behavior.

Step 609, parameter adjustment is performed on the recognition model based on the difference between the recognition result and the actual result, so as to obtain a trained target recognition model.

The execution of steps 606 to 609 may refer to the execution of any embodiment of the present disclosure, and will not be described herein.

According to the model training method for identifying the single-towing operation behavior of the fishing boat, the time characteristics are obtained by inputting the target historical track data into the time flow network of the first network; according to the target historical track data and the time characteristics, adopting a space flow network of the first network to perform characteristic extraction so as to obtain space characteristics; and fusing the time features and the space features to obtain space-time fusion features. Therefore, based on the knowledge learned by different network layers of the first network, the space-time fusion characteristic can be effectively determined.

In order to clearly illustrate how the initial historical track data is filtered to obtain the target historical track data in any of the above embodiments of the present disclosure, the present disclosure further provides a model training method for identifying the behavior of a single-towing operation of a fishing boat.

As shown in fig. 8, the model training method for identifying the behavior of a fishing boat single towing operation may include:

step 801, obtaining initial historical track data of a sample fishing boat with single-towing operation behaviors.

The implementation of step 801 may refer to the implementation of any embodiment of the present disclosure, which is not described herein.

At step 802, correlation analysis is performed on the initial historical track data to identify target historical track data from the initial historical track data.

As one possible implementation, the correlation analysis may include at least one of a spatial correlation analysis, a velocity correlation analysis, a temporal correlation analysis, a heading correlation analysis, a convex hull correlation analysis, a loop correlation analysis, an auto-intersection correlation analysis.

As an example:

1. In the case that the initial historical track data comprises longitudes and latitudes of a plurality of track points, the initial historical track data is subjected to spatial correlation analysis

When the space correlation analysis is carried out on the initial historical track data, the longitude and latitude corresponding to the track points with continuous time sequence and the first set number can be obtained; the spatial distance between the track points can be determined based on the longitude and latitude corresponding to the track points, and the initial historical track data can be screened based on the spatial distance between the track points; it should be noted that, the value of the first set number is not limited in this disclosure.

For example, assume that the set number is N ₁ ，N ₁ Track point p among track points _i Longitude and latitude of (a)<lon _i ,lat _i >，i∈[1,N ₁ ]And i is a positive integer, the locus point p _j Longitude and latitude of (a)<lon _j ,lat _j >，j∈[1,N ₁ ]And j is a positive integer, i is not equal to j; two locus of points p may be employed _i And p _j The euclidean distance between the two track points measures the spatial distance between the two track points, namely the spatial distance S (i, j) between the two track points can be determined according to the following formula:

；(16)

wherein lon _i Represents the ith trace point p _i Longitude, lat of (1) _i Represents the ith trace point p _i Latitude of (a); lon (lon) _j Represents the j-th locus point p _j Longitude, lat of (1) _j Represent the firstj locus points p _j Is a latitude of (c).

Thus, can be based on N ₁ And screening the initial historical track data according to the space distance between each track point in each track point. For example, can be applied to N ₁ Carrying out weighted average on the space distances among all the track points in the track points to obtain a first average value; and comparing the space distance between the track points with the first mean value, and deleting the initial historical track data of the track points corresponding to the target space distance when the target space distance is larger than the first mean value in the space distance between the track points.

The above-mentioned method is based on N ₁ The example of screening the initial historical track data is merely exemplary, and may be other in practical applications, which is not limited by the present disclosure.

2. In the case that the initial historical track data comprises the navigational speeds of a plurality of track points, the initial historical track data is subjected to speed correlation analysis

When the initial historical track data is subjected to speed correlation analysis, the navigational speed corresponding to the track points with continuous time sequences and the second set number can be obtained; the navigation speed distance between the track points can be determined based on the navigation speed corresponding to the track points, and the initial historical track data can be screened based on the navigation speed distance between the track points; it should be noted that, the value of the second set number is not limited in this disclosure.

For example, assume that the set number is N ₂ ，N ₂ Track point p among track points _i Is speed of speed _i ，i∈[1,N ₂ ]And i is a positive integer, the locus point p _j Is speed of speed _j ，j∈[1,N ₂ ]And j is a positive integer, i is not equal to j; two locus of points p may be employed _i And p _j The absolute value of the velocity difference between the two track points is used for measuring the navigational speed distance between the two track points, namely, the navigational speed distance V (i, j) between the two track points can be determined according to the following formula:

；（17）

thus, can be based on N ₂ And screening the initial historical track data according to the navigational speed distance between each track point in each track point. For example, can be applied to N ₂ Carrying out weighted average on navigational speed distances among all the track points in the track points to obtain a second average value; comparing the navigational speed distance between the track points with the second average value, and deleting the initial historical track data of the track points corresponding to the target navigational speed distance when the target navigational speed distance is larger than the second average value in the navigational speed distance between the track points.

The above-mentioned method is based on N ₂ Examples of screening the initial historical track data are merely exemplary, and may be other in practical applications, which are not limited by the present disclosure.

3. In the case that the initial historical track data comprises the time corresponding to a plurality of track points, performing time correlation analysis on the initial historical track data

When the time correlation analysis is carried out on the initial historical track data, the time corresponding to the track points with continuous time sequence and the third set number can be obtained; the time distance between the track points can be determined based on the time corresponding to the track points, and the initial historical track data can be screened based on the time distance between the track points; note that, the value of the third set number is not limited in this disclosure.

For example, assume that the set number is N ₃ ，N ₃ Track point p among track points _i Time of (5) is time _i ，i∈[1,N ₃ ]And i is a positive integer, the locus point p _j Is the speed of the ship _j ，j∈[1,N ₃ ]And j is a positive integer, i is not equal to j; two locus of points p may be employed _i And p _j The absolute value of the time difference between the two track points is used for measuring the time distance between the two track points, namely, the time distance T (i, j) between the two track points can be determined according to the following formula:

；（18）

thus, can be based on N ₃ And screening the initial historical track data according to the time distance between each track point in each track point. For example, can be applied to N ₃ The time distances among all the track points in the track points are weighted and averaged to obtain a third average value; and comparing the time distance between the track points with the third average value, and deleting the initial historical track data of the track points corresponding to the target time distance when the target time distance is larger than the third average value in the time distance between the track points.

The above-mentioned method is based on N ₃ The example of screening the initial historical track data is merely exemplary, and may be other in practical applications, which is not limited by the present disclosure.

4. Under the condition that the initial historical track data comprises the heading corresponding to a plurality of track points, performing time correlation analysis on the initial historical track data

When course correlation analysis is carried out on the initial historical track data, the course corresponding to the track points with continuous time sequence and the fourth set number can be obtained; the course distance between the track points can be determined based on the course corresponding to the track points, and the initial historical track data can be screened based on the course distance between the track points; note that, the value of the fourth set number is not limited in this disclosure.

For example, assume that the set number is N ₄ ，N ₄ Track point p among track points _i Is the delay of _i ，i∈[1,N ₄ ]And i is a positive integer, the locus point p _j Is the navigational speed of degree _j ，j∈[1,N ₄ ]And j is a positive integer, i is not equal to j; two locus of points p may be employed _i And p _j The course angle between the two track points is used for measuring the course distance between the two track points, namely the course distance D (i, j) between the two track points can be determined according to the following formula:

；（19）

Thus, can be based on N ₄ And screening the initial historical track data according to the course distance between each track point in each track point. For example, can be applied to N ₄ Carrying out weighted average on the course distances among all the track points in the track points to obtain a fourth average value; and comparing the heading distance between the track points with a fourth mean value, and deleting initial historical track data of the track points corresponding to the target heading distance when the target heading distance is larger than the fourth mean value in the heading distance between the track points.

The above-mentioned method is based on N ₄ The examples of screening the initial historical track data for the course distance between each track point in the track points are merely exemplary, and may be other in practical applications, which is not limited in this disclosure.

5. Performing convex hull correlation analysis on the initial historical track data under the condition that the initial historical track data comprises longitudes and latitudes of a plurality of track points and corresponding time

When the convex hull correlation analysis is carried out on the initial historical track data, the longitude and latitude and the corresponding time of a fifth set number of track points with continuous time sequence can be obtained; according to longitude and latitude and corresponding time of the track points, a convex hull algorithm (Convex Hull Algorithm) can be adopted to determine convex polygons constructed by the track points; the minimum circumscribed rectangle of the convex polygon is determined by adopting a rotary shell-clamping algorithm, and the aspect ratio of the minimum circumscribed rectangle can be determined.

Therefore, the initial historical track data can be screened based on the aspect ratio of the minimum circumscribed rectangle of the convex polygon determined by the track points of the fifth set number. For example, when the aspect ratio of the minimum circumscribed rectangle is greater than the first set threshold, it indicates that the initial historical track data corresponding to the fifth set number of track points of the sample fishing boat is data during the single-towing operation of the sample fishing boat, and at this time, the initial historical track data corresponding to the fifth set number of track points can be reserved.

It should be noted that the first set threshold may be preset, for example, may be 4, 5, etc., which is not limited in this disclosure.

6. In the case that the initial historical track data comprises longitude and latitude of a plurality of track points and corresponding time, performing loop correlation analysis on the initial historical track data

When the initial historical track data is subjected to ring correlation analysis, the longitude and latitude of a sixth set number of track points with continuous time sequences and corresponding time can be obtained; for any track point A in the sixth set number of track points, a target track point B with a time interval exceeding a set duration (such as 1.5h, 2h and the like) with the track point A can be determined from the track points based on the time corresponding to each track point; based on the longitude and latitude of the track point A and the longitude and latitude of the target track point B, a GeoHash algorithm can be adopted to determine a GeoHash code corresponding to the track point A and a GeoHash code corresponding to the target track point B, and the GeoHash distance between the track point A and the target track point B can be determined according to the GeoHash code corresponding to the track point A and the GeoHash code corresponding to the target track point B; judging whether the GeoHash distance between the track point A and the target track point B is larger than a second set threshold value or not; when the GeoHash distance between the track point a and the target track point B is smaller than the second set threshold, it is determined that a closed loop can be formed between the track point a and the target track point B.

Note that, the value of the sixth set number is not limited in this disclosure.

It should be further noted that the present disclosure does not limit the value of the second set threshold.

Thus, the initial historical track data can be screened based on the result of whether or not a closed loop can be formed between any one of the track points of the sixth set number and the track points adjacent in time sequence. Still further to the above description, for any one of the track points a, when a closed loop may be formed between the track point a and the target track point B corresponding to the track point a, it is indicated that the initial historical track data corresponding to the track point a of the sample fishing boat is data during the single-towing operation of the sample fishing boat, and at this time, the initial historical track data corresponding to the track point a may be retained.

7. Where the initial historical track data includes the longitude and latitude of the plurality of track points and the corresponding time, performing an auto-intersection correlation analysis on the initial historical track data

When the self-intersection correlation analysis is carried out on the initial historical track data, the longitude and latitude and the corresponding time of a seventh set number of track points with continuous time sequences can be obtained; based on the longitude and latitude of each track point and the corresponding time, a Self-interaction algorithm can be adopted to determine whether each track point is intersected automatically in the corresponding time period; the starting time of the time period may be determined for the time of the track point with the earliest time among the track points, and the ending time of the time period may be determined for the time of the track point with the latest time.

Thus, the initial historical track data can be screened based on the analysis result of the self-intersecting correlation analysis of the initial historical track data. For example, when the analysis result of the self-intersecting correlation analysis on the initial historical track data indicates that each track point does not self-intersect, it indicates that the initial historical track data corresponding to the seventh set number of track points of the sample fishing boat is data when the sample fishing boat is in single-towing operation, at this time, the initial historical track data corresponding to the seventh set number of track points can be reserved.

The value of the seventh set number is not limited in this disclosure.

Step 803, training the recognition model according to the target historical track data to obtain a trained target recognition model.

The execution of step 803 may refer to the execution of any embodiment of the present disclosure, and will not be described herein.

According to the model training method for identifying the single-towing operation behavior of the fishing boat, the initial historical track data is subjected to correlation analysis to identify target historical track data from the initial historical track data. Therefore, the method can effectively acquire the target historical track data based on correlation analysis of the initial historical track data.

The above-described embodiments correspond to a model training method for identifying a single-towing operation behavior of a fishing vessel, and the following embodiments apply a trained model for identifying a single-towing operation behavior of a fishing vessel, i.e., a target identification model.

As shown in fig. 9, the method for identifying the single-towing operation behavior of the fishing boat may include:

step 901, acquiring operation behavior data of a target fishing boat.

In an embodiment of the present disclosure, the target fishing vessel may be a fishing vessel to be identified as having a single-towing operation behavior.

In the embodiment of the disclosure, the operation behavior data of the target fishing vessel can be acquired. For example, the operational behavior data of the target fishing vessel may be obtained by radar, satellite navigation system, AIS equipment, etc.

It should be noted that the above example of acquiring the operational behavior data of the target fishing vessel is merely exemplary, and in practical application, the operational behavior data of the target fishing vessel may also be acquired by other manners, which is not limited in this disclosure.

And step 902, identifying the operation behavior data by adopting a target identification model obtained by a model training method for identifying the single-towing operation behavior of the fishing boat so as to determine whether the target fishing boat has the single-towing operation behavior.

In the embodiment of the disclosure, the model training method for identifying the single-towing operation behavior of the fishing vessel in any of the embodiments may be used to obtain the target identification model.

In the embodiment of the disclosure, the target recognition model can be adopted to recognize the operation behavior data of the target fishing boat, and whether the target fishing boat has the single-towing operation behavior can be determined.

Therefore, by adopting the identification method for the single-towing operation behavior of the fishing boat, the single-towing operation behavior of the fishing boat can be intelligently identified, and relevant staff can be assisted to realize supervision on the fishing boat, so that the problems of illegal fishing of the fishing boat and illegal use of net work can be solved.

According to the identification method of the single-towing operation behavior of the fishing boat, operation behavior data of a target fishing boat are obtained; and identifying the operation behavior data by adopting a target identification model obtained by a model training method for identifying the single-towing operation behavior of the fishing boat so as to determine whether the target fishing boat has the single-towing operation behavior. Thus, the identification of the single-towing operation behavior of the target fishing boat can be realized.

Corresponding to the model training method for identifying the behavior of the single-towing operation of the fishing boat provided by the embodiments of fig. 1 to 8, an embodiment of the present disclosure further provides a model training device for identifying the behavior of the single-towing operation of the fishing boat, and since the model training device for identifying the behavior of the single-towing operation of the fishing boat provided by the embodiments of the present disclosure corresponds to the model training method for identifying the behavior of the single-towing operation of the fishing boat provided by the embodiments of the present disclosure, the implementation of the model training method for identifying the behavior of the single-towing operation of the fishing boat is also applicable to the model training device for identifying the behavior of the single-towing operation of the fishing boat provided by the embodiment, which is not described in detail in the embodiment.

As shown in fig. 10, the model training apparatus 1000 for identifying a fishing boat single drag operation behavior may include an acquisition module 1001, a filtering module 1002, and a training module 1003.

The acquiring module 1001 is configured to acquire initial historical track data of a sample fishing boat with a single-towing operation.

The filtering module 1002 is configured to filter the initial historical track data to obtain target historical track data; the target historical track data are data of the sample fishing boat in single-towing operation.

The training module 1003 is configured to train the recognition model according to the target historical track data, so as to obtain a trained target recognition model; the recognition model is used for recognizing whether the single drag operation acts or not based on the target historical track data.

In one possible implementation of the embodiments of the present disclosure, the training module 1003 is configured to: inputting the target historical track data into a first network of an identification model to obtain space-time fusion characteristics; the first network is used for extracting space-time fusion characteristics; inputting the space-time fusion features into a second network of the recognition model to obtain time sequence features; inputting the time sequence characteristics into an attention network of the recognition model to obtain weighted fusion characteristics; according to the weighted fusion characteristics, an output layer of the identification model is adopted to identify whether the sample fishing boat has a single-towing operation behavior; and carrying out parameter adjustment on the recognition model based on the difference between the recognition result and the actual result so as to obtain a trained target recognition model.

In one possible implementation of an embodiment of the present disclosure, the second network is a Bi-directional long-short-term memory network Bi-LSTM.

In one possible implementation of an embodiment of the present disclosure, the first network includes a time stream network and a space stream network; training module 1003 for: inputting the target historical track data into a time flow network of a first network to obtain a time characteristic; based on the target historical track data, adopting a space flow network and a time flow network of the first network to perform feature extraction so as to obtain space features; and fusing the time features and the space features to obtain space-time fusion features.

In one possible implementation of an embodiment of the disclosure, the time-flow network includes a first feature extraction layer, a second feature extraction layer, a third feature extraction layer, a fourth feature extraction layer, and a fifth feature extraction layer connected in series; the spatial stream network comprises a sixth feature extraction layer, a seventh feature extraction layer, an eighth feature extraction layer, a ninth feature extraction layer, a tenth feature extraction layer, a first fusion feature layer and a second fusion feature layer which are connected in series; training module 1003 for: inputting the target historical track data and the first feature into a sixth feature extraction layer of the spatial stream network to obtain a sixth feature; wherein the first feature is an output of the first feature extraction layer; inputting the sixth feature and the second feature into a seventh feature extraction layer of the spatial stream network to obtain a seventh feature; wherein the second feature is an output of the second feature extraction layer; inputting the seventh feature and the third feature into an eighth feature extraction layer of the spatial stream network to obtain an eighth feature; wherein the third feature is an output of the third feature extraction layer; inputting the eighth feature and the fourth feature into a ninth feature extraction layer of the spatial stream network to obtain a ninth feature; wherein the fourth feature is an output of the fourth feature extraction layer; inputting the ninth feature and the fifth feature into a tenth feature extraction layer of the spatial stream network to obtain a tenth feature; wherein the fifth feature is an output of the fifth feature extraction layer; inputting the tenth feature into a first feature fusion layer of the spatial stream network to obtain a first fusion feature; and inputting the first fusion characteristic into a second characteristic fusion layer of the spatial stream network to obtain the spatial characteristic.

In one possible implementation of the disclosed embodiments, the attention network includes a hidden layer and a fully connected layer; training module 1003 for: inputting the time sequence characteristics into a hidden layer of the attention network to obtain intermediate characteristics; the intermediate features are input into a full connection layer of the attention network to weight the intermediate features, and a weighted fusion feature is obtained.

In one possible implementation of the embodiments of the present disclosure, the filtering module 1002 is configured to: correlation analysis is performed on the initial historical track data to identify target historical track data from the initial historical track data.

In one possible implementation of an embodiment of the present disclosure, the correlation analysis includes at least one of a spatial correlation analysis, a velocity correlation analysis, a temporal correlation analysis, a heading correlation analysis, a convex hull correlation analysis, a loop correlation analysis, an auto-intersection correlation analysis.

In one possible implementation manner of the embodiment of the present disclosure, the model training apparatus 1000 for identifying a fishing boat single towing operation behavior may further include:

the processing module is used for acquiring an operable area list, and screening initial historical track data according to operable areas in the operable area list so as to keep the initial historical track data of the sample fishing boat with the movable range not in the operable area;

Or alternatively, the first and second heat exchangers may be,

and the screening module is used for screening the moving range of the sample fishing boat according to the initial historical track data so as to keep the initial historical track data of the sample fishing boat with the moving range within a set distance range from the coast and/or the port.

The model training device for identifying the single-towing operation behavior of the fishing boat is characterized by acquiring initial historical track data of a sample fishing boat with the single-towing operation behavior; filtering the initial historical track data to obtain target historical track data of the sample fishing boat in single-towing operation; the target historical track data are data of the sample fishing boat in single-towing operation; training the recognition model according to the target historical track data to obtain a trained target recognition model; the recognition model is used for recognizing whether the single drag operation acts or not based on the target historical track data. Therefore, training of the identification model can be achieved, and the prediction effect of the model is improved, namely the accuracy and reliability of the prediction result of the model are improved.

Corresponding to the above-mentioned method for identifying the behavior of a single-drag operation of a fishing boat provided by the embodiment of fig. 9, an embodiment of the present disclosure further provides a device for identifying the behavior of a single-drag operation of a fishing boat, and since the device for identifying the behavior of a single-drag operation of a fishing boat provided by the embodiment of the present disclosure corresponds to the method for identifying the behavior of a single-drag operation of a fishing boat provided by the embodiments of the present disclosure, an implementation of the method for identifying the behavior of a single-drag operation of a fishing boat is also applicable to the device for identifying the behavior of a single-drag operation of a fishing boat provided by the embodiment, which is not described in detail in the embodiment.

As shown in fig. 11, the identification device 1100 for a single drag operation of a fishing boat may include an acquisition module 1101 and an identification module 1102.

The acquisition module 1101 is configured to acquire operation behavior data of the target fishing boat;

the identifying module 1102 is configured to identify the operation behavior data by using the target identifying model obtained by the model training method for identifying a single-towing operation behavior of a fishing boat according to any one of the foregoing embodiments, so as to determine whether the target fishing boat has the single-towing operation behavior.

According to the identification device for the single-towing operation behavior of the fishing boat, operation behavior data of a target fishing boat are obtained; and identifying the operation behavior data by adopting a target identification model obtained by a model training method for identifying the single-towing operation behavior of the fishing boat so as to determine whether the target fishing boat has the single-towing operation behavior. Thus, the identification of the single-towing operation behavior of the target fishing boat can be realized.

According to an embodiment of the disclosure, the disclosure further provides an electronic device.

Wherein, electronic equipment includes:

a processor;

a memory for storing processor-executable instructions;

Wherein the processor is configured to execute instructions to implement a model training method for identifying a fishing vessel single drag operation behavior, or a method of identifying a fishing vessel single drag operation behavior, as set forth in any of the foregoing embodiments.

As an example, fig. 12 is a schematic structural diagram of an electronic device 1200 according to an exemplary embodiment of the present disclosure, and as shown in fig. 12, the electronic device 1200 may further include:

the memory 1210 and the processor 1220 are connected to a bus 1230 of different components (including the memory 1210 and the processor 1220), the memory 1210 stores a computer program, and when the processor 1220 executes the program, the model training method for identifying the behavior of the fishing boat single drag operation or the identification method of the fishing boat single drag operation according to the embodiments of the present disclosure is implemented.

Bus 1230 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus architectures. By way of example, and not limitation, such architectures include Industry Standard Architecture (ISA) bus, micro channel architecture (MAC) bus, enhanced ISA bus, video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnect (PCI) bus.

Electronic device 1200 typically includes a variety of electronic device readable media. Such media can be any available media that is accessible by electronic device 1200 and includes both volatile and nonvolatile media, removable and non-removable media.

Memory 1210 may also include computer system readable media in the form of volatile memory, such as Random Access Memory (RAM) 1240 and/or cache memory 1250. The server 1200 may further include other removable/non-removable, volatile/nonvolatile computer system storage media. By way of example only, the storage system 1260 may be used to read from or write to non-removable, non-volatile magnetic media (not shown in FIG. 12, commonly referred to as a "hard disk drive"). Although not shown in fig. 12, a magnetic disk drive for reading from and writing to a removable non-volatile magnetic disk (e.g., a "floppy disk"), and an optical disk drive for reading from or writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In such cases, each drive may be coupled to bus 1230 through one or more data medium interfaces. Memory 1210 may include at least one program product having a set (e.g., at least one) of program modules configured to carry out the functions of the various embodiments of the disclosure.

Program/utility 1280 having a set (at least one) of program modules 1270 may be stored in, for example, memory 1210, such program modules 1270 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each or some combination of which may include an implementation of a network environment. Program modules 1270 generally perform the functions and/or methods in the embodiments described in this disclosure.

The electronic device 1200 may also communicate with one or more external devices 1290 (e.g., keyboard, pointing device, display 1291, etc.), one or more devices that enable a user to interact with the electronic device 1200, and/or any device (e.g., network card, modem, etc.) that enables the electronic device 1200 to communicate with one or more other computing devices. Such communication may occur through an input/output (I/O) interface 1292. Also, the electronic device 1200 may communicate with one or more networks such as a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, e.g., the internet, through a network adapter 1293. As shown, the network adapter 1293 communicates with other modules of the electronic device 1200 over the bus 1230. It should be appreciated that although not shown, other hardware and/or software modules may be used in connection with electronic device 1200, including, but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, data backup storage systems, and the like.

Processor 1220 executes various functional applications and data processing by executing programs stored in memory 1210.

It should be noted that, the implementation process and the technical principle of the electronic device in this embodiment refer to the foregoing explanation of the model training method for identifying the behavior of the single drag operation of the fishing vessel or the identification method of the behavior of the single drag operation of the fishing vessel in the embodiments of the present disclosure, which are not repeated herein.

In an exemplary embodiment, a computer readable storage medium is also provided, for example a memory comprising instructions executable by a processor of an electronic device to perform the model training method for identifying a fishing vessel single drag operation behavior or the identification method of a fishing vessel single drag operation behavior as set forth in any of the above embodiments. Alternatively, the computer readable storage medium may be ROM, random Access Memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

In an exemplary embodiment, a computer program product is also provided, including a computer program/instruction, where the computer program/instruction, when executed by a processor, implements the model training method for identifying a fishing vessel single towing operation behavior or the identification method for a fishing vessel single towing operation behavior set forth in any one of the above embodiments.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This disclosure is intended to cover any adaptations, uses, or adaptations of the disclosure following the general principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It is to be understood that the present disclosure is not limited to the precise arrangements and instrumentalities shown in the drawings, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A model training method for identifying a single drag operation behavior of a fishing vessel, the method comprising:

acquiring initial historical track data of a sample fishing boat with single-towing operation behaviors;

filtering the initial historical track data to obtain target historical track data; the target historical track data are data of the sample fishing boat in single-towing operation;

Training the recognition model according to the target historical track data to obtain a trained target recognition model; the identification model is used for identifying whether the target historical track data has a single-drag operation behavior or not;

training the recognition model according to the target historical track data to obtain a trained target recognition model, wherein the training comprises the following steps:

inputting the target historical track data into a first network of the identification model to obtain space-time fusion characteristics; the first network is used for extracting space-time fusion characteristics;

inputting the space-time fusion features into a second network of the recognition model to obtain time sequence features;

inputting the time sequence characteristics into an attention network of the identification model to obtain weighted fusion characteristics;

according to the weighted fusion characteristics, an output layer of the identification model is adopted to identify whether the sample fishing boat has a single-towing operation behavior;

based on the difference between the recognition result and the actual result, carrying out parameter adjustment on the recognition model to obtain a trained target recognition model;

the first network comprises a time stream network and a space stream network;

The inputting the target history track data into the first network of the recognition model to obtain a space-time fusion feature comprises the following steps:

inputting the target historical track data into a time flow network of the first network to obtain a time characteristic;

based on the target historical track data, adopting a space flow network and a time flow network of the first network to perform feature extraction so as to obtain space features;

fusing the time features and the space features to obtain the space-time fusion features;

the time flow network comprises a first feature extraction layer, a second feature extraction layer, a third feature extraction layer, a fourth feature extraction layer and a fifth feature extraction layer which are connected in series;

the spatial stream network comprises a sixth feature extraction layer, a seventh feature extraction layer, an eighth feature extraction layer, a ninth feature extraction layer, a tenth feature extraction layer, a first fusion feature layer and a second fusion feature layer which are connected in series;

the step of extracting features by using the spatial stream network and the temporal stream network of the first network based on the target historical track data to obtain spatial features includes:

inputting the target historical track data and the first features into a sixth feature extraction layer of the spatial stream network to obtain sixth features; wherein the first feature is an output of the first feature extraction layer;

Inputting the sixth feature and the second feature into a seventh feature extraction layer of the spatial stream network to obtain a seventh feature; wherein the second feature is an output of the second feature extraction layer;

inputting the seventh feature and the third feature into an eighth feature extraction layer of the spatial stream network to obtain an eighth feature; wherein the third feature is an output of the third feature extraction layer;

inputting the eighth feature and the fourth feature into a ninth feature extraction layer of the spatial stream network to obtain a ninth feature; wherein the fourth feature is an output of the fourth feature extraction layer;

inputting the ninth feature and the fifth feature into a tenth feature extraction layer of the spatial stream network to obtain a tenth feature; wherein the fifth feature is an output of the fifth feature extraction layer;

inputting the tenth feature into a first feature fusion layer of the spatial stream network to obtain a first fusion feature;

and inputting the first fusion characteristic into a second characteristic fusion layer of the spatial stream network to obtain the spatial characteristic.

2. The method of claim 1, wherein the second network is a Bi-directional long-short-term memory network Bi-LSTM.

3. The method of claim 1, wherein the attention network comprises a hidden layer and a fully connected layer;

the inputting the time sequence feature into the attention network of the recognition model to obtain a weighted fusion feature comprises:

inputting the time sequence characteristics into a hidden layer of the attention network to obtain intermediate characteristics;

and inputting the intermediate features into a full connection layer of the attention network to weight the intermediate features so as to obtain the weighted fusion features.

4. The method of claim 1, wherein filtering the initial historical track data to obtain target historical track data comprises:

and performing correlation analysis on the initial historical track data to identify the target historical track data from the initial historical track data.

5. The method of claim 4, wherein the correlation analysis comprises at least one of a spatial correlation analysis, a velocity correlation analysis, a temporal correlation analysis, a heading correlation analysis, a convex hull correlation analysis, a loop correlation analysis, a self-intersection correlation analysis.

6. The method of any one of claims 1-5, wherein after the obtaining of the initial historical trajectory data for the sample fishing vessel with single-towing operation behavior, the method further comprises:

Acquiring an operable area list, and screening the initial historical track data according to the operable area in the operable area list so as to keep the initial historical track data of the sample fishing boat with the movable range not in the operable area;

or alternatively, the first and second heat exchangers may be,

and screening the moving range of the sample fishing boat according to the initial historical track data so as to keep the initial historical track data of the sample fishing boat with the moving range within a set distance range from the coast and/or the harbour.

7. A method for identifying a single drag operation of a fishing vessel, the method comprising:

acquiring operation behavior data of a target fishing boat;

the operational behaviour data is identified using a target identification model obtained by a method as claimed in any one of claims 1 to 6 to determine whether the target fishing vessel has single towing operational behaviour.

8. A model training device for identifying a single drag operation of a fishing vessel, the device comprising:

The training module is used for training the recognition model according to the target historical track data so as to obtain a trained target recognition model; the identification model is used for identifying whether the target historical track data has a single-drag operation behavior or not;

the training module is specifically configured to:

the first network comprises a time stream network and a space stream network; the training module is specifically configured to:

the training module is specifically configured to:

9. The apparatus of claim 8, wherein the second network is a Bi-directional long-short-term memory network Bi-LSTM.

10. The apparatus of claim 8, wherein the attention network comprises a hidden layer and a fully connected layer; the training module is specifically configured to:

11. The device according to claim 8, characterized in that said filtering module is in particular adapted to:

12. The apparatus of claim 11, wherein the correlation analysis comprises at least one of a spatial correlation analysis, a velocity correlation analysis, a temporal correlation analysis, a heading correlation analysis, a convex hull correlation analysis, a loop correlation analysis, a self-intersection correlation analysis.

13. The apparatus according to any one of claims 8-12, wherein the apparatus further comprises:

the processing module is used for acquiring an operable area list, and screening the initial historical track data according to the operable areas in the operable area list so as to keep the initial historical track data of the sample fishing boat with the movable range not in the operable area;

or alternatively, the first and second heat exchangers may be,

14. An identification device for single-towing operation of a fishing vessel, the device comprising:

an identification module for identifying the operational behavior data using a target identification model obtained by the method of any one of claims 1-6 to determine whether the target fishing vessel has single-towed operational behavior.

15. An electronic device, comprising:

memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method according to any one of claims 1-6 or the method according to claim 7 when executing the computer program.

16. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the method according to any one of claims 1-6, or the method according to claim 7.