A data mining method for target ship images in a water environment

Abstract

The application discloses a kind of water environment under target ship image data mining algorithm, it is related to image recognition field, including S1: in water environment by intelligent acquisition camera acquisition target ship image;S2: by image segmentation algorithm to the ship image of collection is first segmented and then finds balanced threshold value;S3: utilize ship image edge point extraction formula, and ship image edge is extracted;S4: to the target ship image of S3 processing is carried out effective feature point extraction using neural network;S5: the feature point of target ship image extracted to S4 is neural network training;S6: utilize the error between target image pixel value and expected pixel value, to modify the connection weight value between target image, so that target image is optimized;S7: output target ship image recognition result.The application can improve the quality of target image, and in a variety of environments can achieve good real-time, effectiveness and robustness.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image recognition, more particularly to a data mining method for target ship images in a water environment. BACKGROUND

[0002] According to the current situation of "scattered, disordered and poor" ship transportation and the problems of blind area and many safety hazards existing in the ship supervision platform, the modern technology relies on information technology means, adopts ship intelligent terminal device application, and builds a ship intelligent supervision and service platform to realize real-time dynamic monitoring of ships in the jurisdiction. On this basis, the ship supervision platform needs to process the ship image in real time, identify the ship posture, speed, ship number trajectory and other core safety data.

[0003] The traditional image recognition algorithm is difficult to effectively extract useful recognition features when applied in a fuzzy water environment, and the recognition effect and efficiency of the traditional method are not good in the case of a large amount of data. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a data mining method for target ship images in a water environment, so as to improve the quality of target images and effectively shorten the recognition time of target ship image processing in a fuzzy water environment.

[0005] In order to achieve the above purpose, the present application adopts the following technical scheme:

[0006] A data mining method for target ship images in a water environment, comprising the following steps:

[0007] S1: collecting target ship images through an intelligent acquisition camera in a water environment;

[0008] S2: performing segmentation and then finding a balanced threshold value on the collected ship images through an image segmentation algorithm;

[0009] S3: extracting the edges of the ship images by using a ship image edge point extraction formula;

[0010] S4: performing effective feature point extraction on the target ship images processed in S3 by using a neural network;

[0011] S5: training the target ship image feature points extracted in S4 by using a neural network;

[0012] S6: modifying the connection weight values between the target images by using the error between the pixel values of the target images and the expected pixel values, so as to optimize the target images;

[0013] S7: outputting the target ship image recognition result.

[0014] More specifically:

[0015] The ship image collected in S1 includes the posture, number and track of the ship.

[0016] S2 specifically includes:

[0017] S2.1: find the maximum gray value Z of the target ship image max and the minimum gray value Z min , calculate the initial threshold value:

[0018] S2.2: calculate In this iterative formula, h k is the number of pixels with a pixel gray value size of k in the target image, T i is the threshold value obtained by the i-th calculation, if T i is not an integer, then take the left integer part;

[0019] S2.3: divide the target ship image into foreground and background images, respectively find their average gray values Z0 and Z1, and find the new threshold value of the target ship image segmentation

[0020] S2.4: if T k =T k+1 , then the obtained threshold value is the threshold value, otherwise, participate in step S2.2 to reiterate the operation.

[0021] Based on the rule that data mining obtains the maximum value at the image edge, the edge point extraction formula adopted in S3 is:

[0022]

[0023] Where f(i,j) represents the coordinates of the image edge.

[0024] S6 includes the following steps:

[0025] S6.1: obtain the mean square error between the pixel value of the preprocessed target image and the expected pixel value through the formula q , where r Q is the expected pixel value, N q is the number of layers of the target image, q is the output layer of the target image recognition, O q is the real pixel value;

[0026] S6.2: adopt the intelligent acquisition camera to shoot, which has the following relationship: Where S represents the step size, O q is a function of I i Therefore, we can get:

[0027] S6.3: a target image recognition algorithm based on data mining is adopted:

[0028] Delta w = eta (1-alpha) D (k) + alpha D (k-1) is used for target image recognition, D (k) represents the negative gradient of the target ship image at k time, D (k-1) represents the negative gradient of the target ship image at k-1 time, eta represents the recognition efficiency, and alpha represents the momentum coefficient, wherein the range of eta and alpha is 0 < eta, alpha < 1.

[0029] S7 comprises:

[0030] S7.1: after processing, N image feature points are obtained, and the scale of each feature point is m i , then the scale information of the N target image feature points forms a feature vector:

[0031] M N = (m1, m2,..., m (N-1) , m N ) ;

[0032] S7.2: the information of the N target image feature points is quickly sorted, and N P values can be obtained, that is: P N = sort (M N , descend) ;

[0033] S7.3: the first L P values are intercepted as the feature vector of the target image;

[0034] S7.4: the first L feature vectors of the target image are extracted and classified, and the image recognition result is obtained by S5 based on the trained target image weight value.

[0035] The advantages of the present application over the prior art are that, compared with the application of the traditional image recognition algorithm in the fuzzy water environment, the present application can effectively shorten the recognition time of the target ship image processing in the fuzzy water environment, improve the quality of the target image, and achieve good real-time performance, effectiveness and robustness in various environments. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 is the flowchart of the present application. DETAILED DESCRIPTION

[0037] The specific embodiments of the present application will be described below in combination with the drawings.

[0038] As shown in the drawings, the present application comprises the following steps: Figure 1

[0039] ​S1: collecting target ship image through intelligent collection camera in water environment; the ship image collected in S1 includes core safety data such as posture, ship number and track of the ship.

[0040] S2: performing segmentation first and then finding balanced threshold value through image segmentation algorithm on the collected ship image.

[0041] S2 specifically includes:

[0042] S2.1: obtaining maximum gray value Z of target ship image max and minimum gray value Z min , calculating initial threshold value:

[0043] S2.2: calculating In the iteration formula, h k is the number of pixels with pixel gray value size k of the target image, T i is the threshold value obtained by the i-th calculation, if T i is not an integer, then take the integer to the left;

[0044] S2.3: segmenting the target ship image into foreground image and background image, respectively obtaining average gray values Z0 and Z1 thereof, and obtaining new threshold value of target ship image segmentation

[0045] S2.4: if T k =T k+1 , then the obtained threshold value is the threshold value, otherwise, participate in step S2.2 to perform iteration operation again.

[0046] S3: extracting ship image edge by using ship image edge point extraction formula;

[0047] Based on the rule that data mining obtains maximum value at the image edge, the edge point extraction formula adopted in S3 is:

[0048]

[0049] Wherein f(i,j) represents the coordinates of the image edge.

[0050] S4: performing effective feature point extraction on the target ship image processed in S3 by using neural network;

[0051] S5: performing neural network training on the target ship image feature points extracted in S4;

[0052] S6: using the error between the target image pixel value and the expected pixel value to modify the connection weight value between the target images, so as to optimize the target images;

[0053] S6 includes the following steps:

[0054] S6.1: Obtain the mean square error between the pre-processed target image pixel value and the expected pixel value by the formula , wherein r q is the expected pixel value, N Q is the number of target image layers, q is the output layer of target image recognition, O q is the true pixel value;

[0055] S6.2: Use the intelligent acquisition camera to shoot, which has the following relationship: , wherein S represents the step size, O q is a function of I q . Therefore, we can obtain:

[0056] S6.3: Use the target image recognition algorithm based on data mining:

[0057] Δw = η(1-α)D(k) + αD(k-1) to recognize the target image, D(k) represents the negative gradient of the target ship image at time k, D(k-1) represents the negative gradient of the target ship image at time k-1, η represents the recognition efficiency, and α represents the momentum coefficient, wherein the range of η and α is 0 < η, α < 1.

[0058] S7: Output the target ship image recognition result.

[0059] S7 specifically includes:

[0060] S7.1: Obtain N image feature points after processing, and the scale of each feature point is m i , then the scale information of the N target image feature points forms a feature vector:

[0061] M N = (m1, m2, ···, m (N-1) , m N );

[0062] S7.2: Sort the information of the N target image feature points quickly to obtain N P values, i.e., P N = sort(M N , descend);

[0063] S7.3: Take the first L P values as the feature vector of the target image;

[0064] S7.4: Extract and classify the first L feature vectors of the target image, and convert to S5 to perform image recognition based on the trained target image weight to obtain the image recognition result.

[0065] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art, according to the technical solution and inventive concept of the present application, makes equivalent replacement or change within the technical range disclosed by the present application, should be covered within the protection scope of the present application.

Claims

1. A data mining method for target ship image in a water environment, characterized in that, It comprises the following steps: S1: collecting target ship image through intelligent collection camera in water environment; S2: carrying out segmentation first and then searching for balanced threshold through image segmentation algorithm on collected ship image; S3: using ship image edge point extraction formula to extract ship image edge; S4: adopting neural network to extract effective feature point of target ship image after S3 processing; S5: carrying out neural network training on extracted target ship image feature point of S4; S6: using error between target image pixel value and expected pixel value to modify connection weight value between target images, so that target image is optimized; S7: outputting target ship image recognition result; S6 comprises the following steps: S6.1: Obtain the mean square error between the pre-processed target image pixel value and the expected pixel value by the formula where r q is the expected pixel value, N Q is the number of target image layers, q is the output layer of target image recognition, O q is the real pixel value; S6.2: Adopting intelligent acquisition camera to shoot has the following relationship: In the formula, S represents the step length, O q is the function of I q Therefore, we can get: S6.3: adopting target image recognition algorithm based on data mining: Δw=η(1-α)D(k)+αD(k-1) recognizes target image, D(k) represents negative gradient of target ship image at k moment, D(k-1) represents negative gradient of target ship image at k-1 moment, η represents recognition efficiency, and α represents momentum coefficient, wherein, the range of η and α is 0<η, α<1.

2. The data mining method for target ship image in water environment according to claim 1, characterized in that, The ship image collected in S1 comprises ship posture, ship number and track.

3. The data mining method of the target ship image in the water environment according to claim 1, wherein, S2 specifically comprises: S2.1: Find the maximum gray value Z of the target ship image max and the minimum gray value Z min , calculate the initial threshold value: S2.2: Calculate In this iteration formula, h k The number of pixels with the pixel gray value size k in the target image, T i The threshold value obtained by the i-th calculation, if T i is not an integer, then take the integer part to the left S2.3: segment the target ship image into foreground and background images, respectively find their average gray values Z0 and Z1; find the new threshold value of the target ship image segmentation S2.4: If T k = T k+1 , then the threshold value is obtained, otherwise the iteration is repeated with step S2.

2.

4. The data mining method of the target ship image in the water environment according to claim 1, wherein, Based on the rule that data mining obtains maximum value at image edge, the edge point extraction formula adopted in S3 is: Wherein f(i,j) represents coordinates of image edge.

5. The data mining method for target ship image in water environment according to claim 1, characterized in that, S7 comprises: S7.1: After processing, N image feature points are obtained, and the size of each feature point is m i The N target image feature point size information constitutes a feature vector: M N = (m1, m2, ···, m (N-1) ,m N ) ; S7.2: Fast sorting is performed on the information of the N target image feature points, and N P values can be obtained, i.e., P N = sort(M N , descend); S7.3: intercepting the first L P values as feature vector of target image; S7.4: extracting and classifying the first L feature vectors of target image, and converting S5 to carry out image recognition based on trained target image weight value, so as to obtain image recognition result.

Images