A ship collision avoidance method, system, device and storage medium based on improved VGG network lock

Abstract

The present invention discloses a ship collision avoidance method, system, device and storage medium for a ship lock based on an improved VGG network. The method comprises: obtaining surveillance videos of a no-parking zone and a ship lock through a surveillance lens after deformity correction, and extracting frame images from the surveillance videos; demarcating an ROI area near the ship lock on the frame image, and cropping the ROI area into a sensitive area image; performing detection and early warning based on the sensitive area image according to a pre-built background ship distinction model and a ship lock opening and closing distinction model; adaptively extracting features through convolution, introducing a batch normalization operation and adjusting the pooling layer window size to 3×3, and finally using a softmax classifier to complete the construction of the background ship distinction model and the ship lock opening and closing distinction model, thereby improving the accuracy of the model; the present invention has the advantages of not requiring the installation of an anti-collision sensor, high detection accuracy, and can effectively avoid the problem that a ship in front of the ship lock gate is easily collided with the gate when it is closed, causing damage to the gate.

Description

Technical Field

[0001] The present invention relates to a ship collision avoidance method, system, device and storage medium based on an improved VGG network ship lock, and belongs to the technical field of ship safety. Background Art

[0002] A ship lock is a water transport structure that uses water level fluctuations within a box-shaped lock chamber to raise or lower ships. It is a type of navigation structure used to overcome significant water level differences in a waterway, allowing ships to be raised or lowered from one river level to another. Ship locks are often constructed at water conservancy projects on inland waterways and are sometimes necessary to control tidal ranges at river mouths or harbor entrances.

[0003] The ship lock consists of a sluice gate with gates and valves, a lock chamber for vessels, upstream and downstream fairways for guiding vessels into the lock chamber, a water supply system for filling and draining the lock chamber, and the gate and valve opening and closing mechanisms and control systems. The procedure for a ship passing through the lock from the downstream fairway to the upstream is to use the water supply system to level the chamber water level with the downstream fairway, then open the lower sluice gate, allow the vessel to enter the lock chamber, close the lower sluice gate, fill the lock chamber with water until the water level is level with the upstream fairway, open the upper sluice gate, and allow the vessel to enter the upstream fairway. The procedure is reversed for a ship passing from the upstream fairway to the downstream fairway.

[0004] During ship lock operation, ships near the gates in the lock chamber, while waiting to be moored, are prone to drifting with the current and colliding with the gates, causing damage to both the gates and the ships. Lock gate ship collision prevention is a highly economical application in the water conservancy industry. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method, system, device and storage medium for ship collision avoidance based on an improved VGG network lock to solve the problem that when the lock is closed, the front ship is prone to collide with the gate and cause damage to the gate.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0007] In a first aspect, the present invention provides a ship collision avoidance method for ship locks based on an improved VGG network, comprising:

[0008] Obtain surveillance videos of prohibited parking areas and ship locks through the surveillance footage after deformity correction, and extract frame images from the surveillance videos;

[0009] Delineate the ROI region near the ship lock on the framed image, crop the ROI region into a sensitive region image, and pre-process the sensitive region image;

[0010] Detection is performed based on the pre-processed sensitive area image according to the pre-built background ship distinction model and lock opening and closing distinction model;

[0011] Issue early warning based on the detection results of each frame of sensitive area map;

[0012] The construction of the background ship differentiation model and the lock opening and closing differentiation model includes:

[0013] The pre-processed sensitive area images are marked according to the presence or absence of light and ships to construct a dataset;

[0014] Based on the dataset, a background ship distinction model and a lock opening and closing distinction model are constructed through the VGG network.

[0015] Optionally, marking the sensitive area image according to the presence or absence of light and ships includes:

[0016] The images of sensitive areas in the prohibited parking zone are marked as having a ship or no ship. The having a ship category includes having a ship hull during the day and having a ship hull at night, and the no-ship category includes having no ship hull during the day and having no ship hull at night.

[0017] The images of sensitive areas at the lock are marked as lock open class and lock not open class. The lock open class includes lock open during the day and lock open at night, and the lock not open class includes lock closed during the day, lock closed at night, lock half open and half closed during the day, and lock half open and half closed at night.

[0018] Optionally, the constructing of the background ship differentiation model includes:

[0019] The sensitive area images of the no-parking zone in the dataset are used as sample input to improve the VGG network for forward propagation to extract features;

[0020] The extracted features are classified using the softmax function, and the images in the sensitive areas are classified as having ships or not based on the classification results.

[0021] The error value is calculated based on the cross entropy cost function between the classification and actual classification of the sensitive area image;

[0022] Back propagation is performed based on the error value to update the parameters of the VGG network;

[0023] The VGG network with updated parameters is brought into the above steps for iteration until convergence or the preset number of iterations is reached to obtain the background ship discrimination model.

[0024] Optionally, constructing the lock opening and closing differentiation model includes:

[0025] The sensitive area images of the ship lock in the dataset are used as sample input to improve the VGG network for forward propagation to extract features;

[0026] The extracted features are classified using the softmax function, and the sensitive area images are divided into the lock-open class and the lock-not-open class based on the classification results.

[0027] The error value is calculated based on the cross entropy cost function between the classification and actual classification of the sensitive area image;

[0028] Back propagation is performed based on the error value to update the parameters of the VGG network;

[0029] The VGG network with updated parameters is brought into the above steps for iteration until convergence or the preset number of iterations is reached to obtain the lock opening and closing differentiation model.

[0030] Optionally, the initialization of the VGG network includes: setting an initial learning rate of the VGG network to 0.0001, presetting the number of iterations to 100, and decreasing the learning rate to 10% of the original value every 30 iterations.

[0031] Optionally, the VGG network includes a convolutional layer, a pooling layer, an activation function, and a fully connected layer; the sample input improved VGG network performs forward propagation to extract features including:

[0032] The convolution kernel in the convolution layer traverses the sample data with a preset step size to obtain the feature map;

[0033] The feature map is downsampled through the pooling layer to achieve sample data dimensionality reduction, using maximum pooling;

[0034] The Relu activation function is used to perform nonlinear mapping on the sample data after dimensionality reduction and output the hidden feature space;

[0035] The hidden feature space output by the convolution operation, pooling operation and activation function is mapped to the sample label space through the fully connected layer to form a one-dimensional vector of features.

[0036] Optionally, a batch normalization operation is introduced after the convolutional layer to process the feature map, and the processed output is used as the input of the pooling layer, where the pooling layer adopts a window size of 3×3;

[0037] The batch normalization operation:

[0038]

[0039] Among them, y i is the output value of the i-th batch of batch normalization operation, x i is the input value of the i-th batch of batch normalization operation, m is the number of batches, γ and β are the parameters of the VGG network, and β is a preset hyperparameter.

[0040] Optionally, the detecting based on the pre-processed sensitive area image according to a pre-built background ship distinction model and a lock opening and closing distinction model includes:

[0041] Based on the background ship discrimination model, the sensitive area images of the no-parking zone are classified as either with or without ships. Based on the lock opening and closing discrimination model, the sensitive area images of the lock are classified as either open or closed.

[0042] If the sensitive area image of the no-parking zone is of the ship-in-place type, and the sensitive area image at the ship lock is of the ship lock not-open type, then the sensitive area image of the no-parking zone and the sensitive area image at the ship lock are of the warning type.

[0043] Optionally, the issuing of an early warning based on the detection result of each frame of the sensitive area map includes:

[0044] The detection results of the sensitive area images of the no-parking zone and the sensitive area images at the ship lock of the current 2N+1 frames are obtained. If the sensitive area images of the no-parking zone and the sensitive area images at the ship lock for more than N+1 frames are of the warning type, an early warning is issued.

[0045] In a second aspect, an embodiment of the present invention provides a ship collision avoidance system based on an improved VGG network ship lock, the system comprising:

[0046] An image extraction module is used to obtain surveillance videos of no-parking areas and ship locks through the surveillance footage after deformity correction, and to extract frame images from the surveillance videos;

[0047] An image processing module is used to define a ROI region near the ship lock on the frame image, crop the ROI region into a sensitive region image, and pre-process the sensitive region image;

[0048] An image detection module is used to perform detection based on the pre-processed sensitive area image according to a pre-built background ship distinction model and a lock opening and closing distinction model;

[0049] Image warning module, used to issue warnings based on the detection results of each frame of sensitive area map;

[0050] The construction of the background ship differentiation model and the lock opening and closing differentiation model includes:

[0051] The pre-processed sensitive area images are marked according to the presence or absence of light and ships to construct a dataset;

[0052] Based on the data set, a background ship distinction model and a lock opening and closing distinction model are constructed through the preset VGG network.

[0053] In a third aspect, the present invention provides a ship collision avoidance device for ship locks based on an improved VGG network, comprising a processor and a storage medium;

[0054] The storage medium is used to store instructions;

[0055] The processor is configured to operate according to the instructions to execute the steps of any of the above methods.

[0056] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the program implements the steps of any of the above-mentioned methods when executed by a processor.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] The present invention provides a ship collision avoidance method, system, device and storage medium based on an improved VGG network for ship locks. The method adopts video imaging technology and is simpler and easier to implement than sensor solutions that require installation by mariners and underwater construction. The VGG network that introduces batch normalization processing and adjusts the size of the pooling layer has high detection accuracy, and as the sample size increases, this advantage becomes more prominent than traditional image processing. The method can effectively prevent ships in front of the ship lock from colliding with the gate when it is closed, causing damage to the gate. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a flow chart of a ship collision avoidance method based on an improved VGG network lock provided by an embodiment of the present invention;

[0060] Figure 2 This is an overall flow chart of the model building and detection method provided by an embodiment of the present invention;

[0061] Figure 3 This is a hardware connection diagram of a ship collision avoidance system based on an improved VGG network lock provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0063] Example 1:

[0064] like Figure 1-2 As shown, this embodiment provides a ship collision avoidance method based on an improved VGG network lock, comprising the following steps:

[0065] (1) Obtain surveillance videos of prohibited parking areas and ship locks through the surveillance camera after deformity correction, and extract frame images from the surveillance videos;

[0066] Deformity correction of surveillance lenses can eliminate the distortion introduced by deviations in lens manufacturing precision and assembly process and the resulting distortion of the original image. The deformity-corrected frame images can be extracted from the surveillance video through the RTSP protocol.

[0067] (2) Delineating a ROI region near the ship lock on the framed image, cropping the ROI region into a sensitive region image, and preprocessing the sensitive region image; preprocessing the sensitive region image includes scaling the image to 224×224.

[0068] (3) Detection is performed based on the pre-processed sensitive area image according to the pre-built background ship distinction model and the lock opening and closing distinction model;

[0069] Based on the background ship discrimination model, the sensitive area images of the no-parking zone are classified as either with or without ships. Based on the lock opening and closing discrimination model, the sensitive area images of the lock are classified as either open or closed.

[0070] If the sensitive area image of the no-parking zone is of the ship-in-place type, and the sensitive area image at the ship lock is of the ship lock not-open type, then the sensitive area image of the no-parking zone and the sensitive area image at the ship lock are of the warning type.

[0071] (4) Issue early warning based on the detection results of each frame of the sensitive area map;

[0072] The detection results of the sensitive area images of the no-parking zone and the sensitive area images at the ship lock of the current 2N+1 frames are obtained. If the sensitive area images of the no-parking zone and the sensitive area images at the ship lock for more than N+1 frames are of the warning type, an early warning is issued.

[0073] Specifically: The construction of the background ship differentiation model and the lock opening and closing differentiation model includes:

[0074] 1. Label the pre-processed sensitive area images according to the presence or absence of light and ships to construct a dataset;

[0075] 1.1. Images of sensitive areas in prohibited parking zones are marked as either with or without ships. The with-ship category includes both with and without ships during the day, and the without-ship category includes both without and without ships during the day.

[0076] 1.2. The images of sensitive areas at the lock are marked as lock-open and lock-closed. The lock-open category includes lock-open during the day and lock-open at night. The lock-closed category includes lock-close during the day, lock-close at night, lock-partially open during the day, and lock-partially open at night.

[0077] 2. Based on the dataset, a background ship distinction model and a lock opening and closing distinction model are constructed through the VGG network.

[0078] 2.1. Constructing a background ship differentiation model

[0079] 2.1.1. Use the sensitive area images of the no-parking zone in the dataset as sample input to improve the VGG network and perform forward propagation to extract features;

[0080] 2.1.2. Use the softmax function to classify the extracted features and distinguish the sensitive area images as having a ship or not having a ship based on the classification results;

[0081] 2.1.3. Calculate the error value between the classification and actual class based on the sensitive area image using the cross entropy cost function;

[0082] 2.1.4. Update the parameters of the VGG network based on the error value through back propagation;

[0083] 2.1.5. Bring the VGG network with updated parameters into the above steps and iterate until convergence or the preset number of iterations is reached to obtain the background ship discrimination model.

[0084] 2.2. Constructing the lock opening and closing differentiation model includes:

[0085] 2.2.1. The sensitive area images of the ship lock in the dataset are used as sample input to improve the VGG network for forward propagation to extract features;

[0086] 2.2.2. Use the softmax function to classify the extracted features and distinguish the sensitive area images into the lock open class and the lock closed class based on the classification results;

[0087] 2.2.3. Calculate the error value between the classification and actual class based on the sensitive area image through the cross entropy cost function;

[0088] 2.2.4. Update the parameters of the VGG network through back propagation based on the error value;

[0089] 2.2.5. Bring the VGG network with updated parameters into the above steps and iterate until convergence or the preset number of iterations is reached to obtain the lock opening and closing differentiation model.

[0090] In the above construction process, the initialization of the VGG network includes: setting the initial learning rate of the VGG network to 0.0001, the preset number of iterations to 100, and the learning rate decreasing to 10% of the original value every 30 iterations.

[0091] The VGG network includes convolutional layers, pooling layers, activation functions, and fully connected layers. Sample input improves the VGG network for forward propagation and feature extraction, including:

[0092] The convolution kernel in the convolution layer traverses the sample data with a preset step size to obtain the feature map;

[0093] The feature map is downsampled through the pooling layer to achieve sample data dimensionality reduction, using maximum pooling;

[0094] The Relu activation function is used to perform nonlinear mapping on the sample data after dimensionality reduction and output the hidden feature space;

[0095] The hidden feature space output by the convolution operation, pooling operation and activation function is mapped to the sample label space through the fully connected layer to form a one-dimensional vector of features.

[0096] In order to speed up the convergence of the model and alleviate the gradient diffusion problem of the VGG neural network, a batch normalization operation is introduced after the convolutional layer to process the feature map, and the processed output is used as the input of the pooling layer. The pooling layer uses a window size of 3×3;

[0097] Batch normalization operation:

[0098]

[0099] Among them, y i is the output value of the i-th batch of batch normalization operation, x i is the input value of the i-th batch of batch normalization operation, m is the number of batches, γ and β are the parameters of the VGG network, and β is a preset hyperparameter.

[0100] This example builds a background ship differentiation model and verifies:

[0101] Table 1 Sample set division

[0102]

[0103] The accuracy test was performed using the background ship distinction model of the present invention, Hog+SVM, sift+SVM and the original VGG algorithm model, and the results were:

[0104] Table 2 Algorithm performance in detecting prohibited parking areas in ship locks

[0105] Algorithm Model Accuracy Hog+svm 88.7579％ SIFT+svm 96.4922％ VGG 96.6072％ VGG_V1 (batch normalization) 98.8499％ VGG_V2 (MaxPool uses a 3×3 window) 96.4347％ Background ship distinction model 99.2237％

[0106] It can be seen that the background ship differentiation model provided by the present invention has a high accuracy rate.

[0107] Example 2:

[0108] like Figure 3 As shown, an embodiment of the present invention provides a ship collision avoidance system based on an improved VGG network ship lock, the system comprising:

[0109] The image extraction module uses the corrected surveillance footage to capture surveillance video of the no-parking zone and the ship lock, and extracts framed images from the video. Two cameras (Camera 1 and Camera 2) are installed above the lock chamber, one facing the no-parking zone and the other facing the ship lock. The video image processing machine, an industrial computer, is connected via a network (switch).

[0110] An image processing module is used to define a ROI region near the ship lock on the frame image, crop the ROI region into a sensitive region image, and pre-process the sensitive region image;

[0111] An image detection module is used to perform detection based on the pre-processed sensitive area image according to a pre-built background ship distinction model and a lock opening and closing distinction model;

[0112] The image warning module is used to issue warnings based on the detection results of each sensitive area image frame. The video image processing machine sends the processing results to the embedded device, which uses an IO control card. The embedded device ultimately controls the response of the sound and light alarm device.

[0113] The construction of the background ship differentiation model and the lock opening and closing differentiation model includes:

[0114] The pre-processed sensitive area images are marked according to the presence or absence of light and ships to construct a dataset;

[0115] Based on the data set, a background ship distinction model and a lock opening and closing distinction model are constructed through the preset VGG network.

[0116] Table 3 Hardware selection

[0117] equipment quantity model camera 2 DS-2CD7A47EWD-LZS(2.8-12mm)(B) Industrial Computer 1 I3-4130+4G memory+128G hard drive+dual network cards IO control card 1 ioLogik E1214 v1.2.4 Sound and light alarm device 1 ordinary

[0118] Example 3:

[0119] Based on the first embodiment, the embodiment of the present invention further provides a ship collision avoidance device for ship locks based on an improved VGG network, including a processor and a storage medium;

[0120] The storage medium is used to store instructions;

[0121] The processor is configured to operate according to the instructions to execute the steps of the above method.

[0122] Example 4:

[0123] Based on the first embodiment, the embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the above method are implemented.

[0124] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0125] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0126] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0127] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0128] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A ship collision avoidance method based on an improved VGG network lock, characterized in that: include: Obtain surveillance videos of prohibited parking areas and ship locks through the surveillance footage after deformity correction, and extract frame images from the surveillance videos; Delineate the ROI region near the ship lock on the frame image, crop the ROI region into the sensitive region image, and pre-process the sensitive region image; Detection is performed based on the pre-processed sensitive area image according to the pre-built background ship distinction model and lock opening and closing distinction model; The sensitive area images of the prohibited parking zone are determined as having a ship or not having a ship based on the background ship discrimination model. The sensitive area images of the ship lock are determined as having an open lock or not having an open lock based on the ship lock opening and closing discrimination model. If the sensitive area images of the prohibited parking zone are having a ship and the sensitive area images of the ship lock are not having an open lock, then the sensitive area images of the prohibited parking zone and the sensitive area images of the ship lock are classified as warnings. Issue early warning based on the detection results of each frame of sensitive area map; The construction of the background ship differentiation model and the lock opening and closing differentiation model includes: The pre-processed sensitive area images are marked according to the presence or absence of light and ships to construct a dataset; Based on the dataset, a background ship distinction model and a lock opening and closing distinction model are constructed through the VGG network.

2. The ship collision avoidance method based on improved VGG network lock according to claim 1 is characterized in that: The marking of the sensitive area image according to the presence or absence of light and ships includes: The images of sensitive areas in the prohibited parking zone are marked as having a ship or no ship. The having a ship category includes having a ship hull during the day and having a ship hull at night, and the no-ship category includes having no ship hull during the day and having no ship hull at night. The images of sensitive areas at the lock are marked as lock open class and lock not open class. The lock open class includes lock open during the day and lock open at night, and the lock not open class includes lock closed during the day, lock closed at night, lock half open and half closed during the day, and lock half open and half closed at night.

3. The ship collision avoidance method based on improved VGG network lock according to claim 2 is characterized in that: The constructing of the background ship differentiation model comprises: The sensitive area images of the no-parking zone in the dataset are used as sample input to improve the VGG network for forward propagation to extract features; The extracted features are classified using the softmax function, and the images in the sensitive areas are classified as having ships or not based on the classification results. The error value is calculated based on the cross entropy cost function between the classification and actual classification of the sensitive area image; Back propagation is performed based on the error value to update the parameters of the VGG network; The VGG network with updated parameters is brought into the above steps for iteration until convergence or the preset number of iterations is reached to obtain the background ship discrimination model.

4. The ship collision avoidance method based on improved VGG network lock according to claim 2 is characterized in that: The construction of the ship lock opening and closing differentiation model includes: The sensitive area images of the ship lock in the dataset are used as sample input to improve the VGG network for forward propagation to extract features; The extracted features are classified using the softmax function, and the sensitive area images are divided into the lock-open class and the lock-not-open class based on the classification results. The error value is calculated based on the cross entropy cost function between the classification and actual classification of the sensitive area image; Back propagation is performed based on the error value to update the parameters of the VGG network; The VGG network with updated parameters is brought into the above steps for iteration until convergence or the preset number of iterations is reached to obtain the lock opening and closing differentiation model.

5. A ship collision avoidance method based on an improved VGG network lock according to claim 3 or 4, characterized in that: The initialization of the VGG network includes: setting the initial learning rate of the VGG network to 0.0001, presetting the number of iterations to 100, and decreasing the learning rate to 10% of the original value every 30 iterations.

6. A ship collision avoidance method based on an improved VGG network lock according to claim 3 or 4, characterized in that: The VGG network includes a convolutional layer, a pooling layer, an activation function, and a fully connected layer; The sample input improves the VGG network to perform forward propagation to extract features, including: The convolution kernel in the convolution layer traverses the sample data with a preset step size to obtain the feature map; The feature map is downsampled through the pooling layer to achieve sample data dimensionality reduction, using maximum pooling; The Relu activation function is used to perform nonlinear mapping on the sample data after dimensionality reduction and output the hidden feature space; The hidden feature space output by the convolution operation, pooling operation and activation function is mapped to the sample label space through the fully connected layer to form a one-dimensional vector of features.

7. The ship collision avoidance method based on improved VGG network lock according to claim 6 is characterized in that: A batch normalization operation is introduced after the convolutional layer to process the feature map, and the processed output is used as the input of the pooling layer, and the pooling layer adopts a window size of 3×3; The batch normalization operation: ； in, is the first batch normalization operation The output value of the batch, is the first batch normalization operation batch of input values, is the number of batch processing, and are the parameters of the VGG network, are the preset hyperparameters.

8. The ship collision avoidance method based on improved VGG network lock according to claim 1 is characterized in that: The early warning according to the detection result of each frame of the sensitive area map includes: Get Current The detection results of the sensitive area images of the no-parking zone and the sensitive area images of the lock are as follows: If the sensitive area images of the no-parking zone and the sensitive area images at the ship lock in the frame are of warning type, an early warning is issued.

9. A ship collision avoidance system based on an improved VGG network lock, characterized in that: The system comprises: An image extraction module is used to obtain surveillance videos of no-parking areas and ship locks through the surveillance footage after deformity correction, and to extract frame images from the surveillance videos; An image processing module is used to define a ROI region near the ship lock on the frame image, crop the ROI region into a sensitive region image, and pre-process the sensitive region image; An image detection module is configured to perform detection based on the preprocessed sensitive area image according to a pre-built background ship discrimination model and a lock opening and closing discrimination model; the sensitive area image of the prohibited parking zone is determined as having a ship or no ship according to the background ship discrimination model, and the sensitive area image at the ship lock is determined as having an open lock or not open lock according to the ship lock opening and closing discrimination model; if the sensitive area image of the prohibited parking zone is having a ship and the sensitive area image at the ship lock is not open, then the sensitive area images of the prohibited parking zone and the sensitive area images at the ship lock are classified as warning; Image warning module, used to issue warnings based on the detection results of each frame of sensitive area map; The construction of the background ship differentiation model and the lock opening and closing differentiation model includes: The pre-processed sensitive area images are marked according to the presence or absence of light and ships to construct a dataset; Based on the data set, a background ship distinction model and a lock opening and closing distinction model are constructed through the preset VGG network.

10. A ship collision avoidance device based on an improved VGG network ship lock, characterized in that: including processor and storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images