Intelligent decision-making method and system for unmanned surface vehicle

1. An intelligent decision-making system comprises a processor and a memory, the memory is configured to store a computer-executable instruction, and when the intelligent decision-making system runs, the processor executes the computer-executable instruction in the memory, to perform following operation steps:

acquire an image sequence signal of an observation region;

receive the image sequence signal, recognize a preset interested object appeared in a scenario, determine and identify an image region occupied by the preset interested object to be recognized at water surface, distinguish to which classification the interested object belongs by a convolutional neural network model simultaneously, screen out the interested object that causes a threat to an unmanned platform itself according to the classification of the interested object, continuously observe the interested object, and generate tracking trajectory information of the interested object;

calculate an object threat factor according to the tracking trajectory information, an optical axis angle, the classification, and an attribute of the interested object, wherein the object threat factor comprises an angle threat factor, a speed threat factor, a height threat factor, a distance threat factor and an object classification factor; and

evaluate a threat classification according to a number of the threat factors and values of the threat factors, give a threat level, and lock the interested object with a high threat level for tracking in real time.

2. The intelligent decision-making system for the unmanned surface vehicle according to claim 1 , wherein when determining the image region occupied by the preset interested object, the image region is identified by a rectangular frame, and a central position of the rectangular frame is taken as an image coordinate of the interested object.

3. The intelligent decision-making system for the unmanned surface vehicle according to claim 1 , wherein the image sequence signal comprises a sequence signal of a visible light image and a sequence signal of an infrared image.

4. The intelligent decision-making system for the unmanned surface vehicle according to claim 1 , wherein the convolutional neural network model is specifically generated by the following steps:

using environmental information including a visible light image, an infrared image, an object navigation trajectory, and a position of the unmanned surface vehicle as an input of a convolutional layer;

obtaining a high-layer convolution feature map by continuously abstracting and combining input image features through a convolution template of the convolutional layer;

by a pooling layer, performing an aggregation statistic on the high-layer convolution feature map output by the convolutional layer to reduce a dimension of a feature vector, and outputting an aggregation value of an object feature map;

inputting the aggregation value of the object feature map output by the pooling layer into a fully connected layer of the convolutional neural network to output an object classification eigenvalue; and

taking the object classification eigenvalue output by the fully connected layer as an input value of a back propagation, and correcting the convolution template of the convolutional layer and a weight value of the fully connected layer by using a gradient descent algorithm to complete a training process for the convolution neural network once.

5. The intelligent decision-making system for the unmanned surface vehicle according to claim 4 , wherein a weight value of the high-layer convolutional feature map is updated for a second time, an infrared feature and a convolution template for the infrared feature are introduced into the convolutional layer, an infrared feature weight matrix is obtained by multiplying an infrared eigenvalue and the convolution template for the infrared feature, a new convolutional feature map is obtained by multiplying the infrared feature weight matrix and an element corresponding to the convolutional feature map obtained by a convolution template for a feature extraction of an image pixel matrix, an image feature map is obtained by adding the new convolutional feature map and a feature map obtained with convolution by a convolution template for a feature extraction of another image pixel matrix.

6. An intelligent decision-making method for an unmanned surface vehicle, comprising:

acquiring an image sequence signal of an observation region;

recognizing a preset interested object on water surface appeared in a scenario based on the image sequence signal, determining and identifying an image region occupied by the preset interested object, distinguishing a classification of the interested object by a convolutional neural network model simultaneously, screening out the interested object that causes a threat to an unmanned platform itself according to the classification of the interested object, continuously observing the interested object, and generating tracking trajectory information of the interested object;

calculating an object threat factor according to the tracking trajectory information, an optical axis angle, the classification, and an attribute of the interested object, wherein the interested object threat factor comprises an angle threat factor, a speed threat factor, a height threat factor, a distance threat factor and an object classification factor; and

evaluating a threat classification according to a number of the threat factors and values of the threat factors, giving a threat level, and locking the interested object with a high threat level for tracking in real time.

7. The intelligent decision-making method for the unmanned surface vehicle according to claim 6 , wherein when determining the image region occupied by the preset interested object, the image region is identified by a rectangular frame, and a central position of the rectangular frame is taken as an image coordinate of the interested object.

8. The intelligent decision-making method for the unmanned surface vehicle according to claim 6 , wherein the image sequence signal comprises a sequence signal of a visible light image and a sequence signal of an infrared image.

9. The intelligent decision-making method for the unmanned surface vehicle according to claim 6 , wherein the convolutional neural network model is specifically generated by the following steps:

using environmental information including a visible light image, an infrared image, an object navigation trajectory, and a position of the unmanned surface vehicle as an input of a convolutional layer;

obtaining a high-layer convolution feature map by continuously abstracting and combining input image features through a convolution template of the convolutional layer;

by a pooling layer, performing an aggregation statistic on the high-layer convolution feature map output by the convolutional layer to reduce a dimension of a feature vector, and outputting an aggregation value of an object feature map;

inputting the aggregation value of the object feature map output by the pooling layer into a fully connected layer of the convolutional neural network to output an object classification eigenvalue; and

taking the object classification eigenvalue output by the fully connected layer as an input value of a back propagation, and correcting the convolution template of the convolutional layer and a weight value of the fully connected layer by using a gradient descent algorithm to complete a training process for the convolution neural network once.

10. The intelligent decision-making method for the unmanned surface vehicle according to claim 9 , wherein the method further comprises the following steps:

updating a weight value of the high-layer convolutional feature map for a second time,

introducing an infrared feature and a convolution template for the infrared feature into the convolutional layer,

obtaining an infrared feature weight matrix by multiplying an infrared eigenvalue and the convolution template for the infrared feature,

obtaining a new convolutional feature map by multiplying the infrared feature weight matrix and an element corresponding to the convolutional feature map obtained by a convolution template for a feature extraction of an image pixel matrix,

obtaining an image feature map by adding the new convolutional feature map and a feature map obtained with convolution by a convolution template for a feature extraction of another image pixel matrix.