Port container layout detection method, device, equipment and readable storage medium

Abstract

The present invention discloses a method, device, equipment and readable storage medium for detecting the layout of a port container. The method comprises: acquiring a remote sensing image, preprocessing the remote sensing image, inputting the preprocessed remote sensing image into a preset semantic segmentation model to obtain a semantic image, inputting the semantic image into a preset target detection model to obtain single container information in the semantic image, and outputting the layout information of the port container based on the single container information. The method solves the technical problem of low efficiency of port container dispatching in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of neural network technology in financial technology, and in particular to a port container layout detection method, device, equipment and readable storage medium. Background Art

[0002] With the continuous development of financial technology, especially Internet technology finance, more and more technologies (such as distributed, blockchain, artificial intelligence, etc.) are applied in the financial field, but the financial industry also puts forward higher requirements on technology, such as higher requirements on the distribution of corresponding to-do items in the financial industry.

[0003] At present, with the development of China's ports, the port container transportation industry is also growing stronger and stronger, and the number of containers in a single port is also increasing. How to efficiently dispatch containers has become the key to improving port transportation efficiency. The premise of efficient dispatching containers is to fully control the layout of port containers. At present, video surveillance is usually used in ports to obtain the status of port containers in real time, but this method cannot accurately locate the position and type of containers, that is, it is impossible to fully control the layout of port containers, and thus it is difficult to efficiently dispatch containers, which reduces port transportation efficiency. Therefore, there is a technical problem of low port container dispatching efficiency in the existing technology. Summary of the invention

[0004] The main purpose of the present invention is to provide a port container layout detection method, device, equipment and readable storage medium, aiming to solve the technical problem of low port container scheduling efficiency in the prior art.

[0005] To achieve the above-mentioned purpose, an embodiment of the present invention provides a port container layout detection method, which is applied to a port container layout detection device, and the port container layout detection method includes:

[0006] Acquire remote sensing images and preprocess the remote sensing images;

[0007] Inputting the preprocessed remote sensing image into a preset semantic segmentation model to obtain a semantic image;

[0008] Inputting the semantic image into a preset target detection model to obtain information about a single container in the semantic image;

[0009] Based on the individual container information, port container layout information is output.

[0010] Optionally, the single container information includes the container location coordinates and the container tilt angle.

[0011] The step of inputting the semantic image into a preset target detection model to obtain the single container information in the semantic image comprises:

[0012] Inputting the semantic image into a preset detection model to select a candidate area in the semantic image and obtain a rectangular target box corresponding to the candidate area;

[0013] Inputting the semantic image into a convolutional neural network model to obtain a plurality of feature maps corresponding to the rectangular target box;

[0014] Performing full connection on the multiple feature graphs to obtain the location coordinates of the container;

[0015] The tilt angle of the container is calculated based on the container position coordinates.

[0016] Optionally, the step of inputting the semantic image into a convolutional neural network model to obtain a plurality of feature maps corresponding to the rectangular target box includes:

[0017] Inputting the semantic image into a convolutional neural network model, obtaining a plurality of preset container features for the rectangular target frame in the convolutional neural network model and weight matrices corresponding to the plurality of preset container features respectively;

[0018] Based on the multiple preset container features and the weight matrix, performing convolution processing on the semantic image to obtain a convolution processing result;

[0019] Performing pooling processing on the convolution processing result to obtain a pooling processing result;

[0020] Based on a preset number of alternating convolution and pooling times, the pooling processing result is again subjected to a corresponding number of alternating convolution and pooling processes to obtain a plurality of feature maps corresponding to the rectangular target box.

[0021] Optionally, the step of acquiring remote sensing images and preprocessing the remote sensing images includes:

[0022] Acquire a preset first initial model and a preset first training data set, wherein the preset first training data includes a first training input image and a first theoretical output result;

[0023] Inputting the first training input image into the preset first initial model to encode the first training input image to obtain an encoding result;

[0024] Decoding the encoding result to obtain a first actual output result;

[0025] Comparing the first actual output result with the first theoretical output result to obtain a first model loss value;

[0026] Based on the first model loss value, the preset first initial model is adjusted and retrained until the first model loss value is less than the preset first model threshold, so as to obtain the preset semantic segmentation model.

[0027] Optionally, the step of encoding the first training input image to obtain an encoding result includes:

[0028] Inputting the first training input image into a preset initial convolutional neural network to obtain high-level semantic information;

[0029] Based on the high-level semantic information, classify and identify the pixels in the first training input image to obtain a recognition and classification result;

[0030] Based on the recognition and classification results, an encoding result is output.

[0031] Optionally, the step of acquiring remote sensing images and preprocessing the remote sensing images includes:

[0032] Acquire a preset second initial model and a preset second training data set, wherein the preset second training data includes a second initial training image and a second theoretical output result;

[0033] Inputting the second initial training image into the preset second initial model to select a candidate area in the second initial training image and obtain a rectangular target box corresponding to the candidate area;

[0034] Performing a preset number of convolution and pooling alternating processes on each of the second initial training images to obtain a plurality of feature maps corresponding to each of the rectangular target frames;

[0035] Fully connect multiple feature maps corresponding to the rectangular target box to obtain a second actual output result;

[0036] Comparing the second actual output result with the second theoretical output result to obtain a second model loss value;

[0037] Based on the second model loss value, the preset second initial model is adjusted and retrained until the second model loss value is less than the preset second model threshold, so as to obtain the preset target detection model.

[0038] Optionally, the step of obtaining a preset second training data set includes:

[0039] Acquire a preset training remote sensing image, and based on a preset data set standard, save the preset training remote sensing image as a preset text format image, and read the preset text format image to obtain the second theoretical output result;

[0040] Performing cropping, data enhancement and normalization processing on the preset text format image to obtain a standard image;

[0041] The standard image is subjected to a mean removal process and converted into a preset data format to obtain the second initial training image.

[0042] The present invention also provides a port container layout detection device, which is applied to a port container layout detection device, and the port container layout detection device comprises:

[0043] A preprocessing module, used for acquiring remote sensing images and preprocessing the remote sensing images;

[0044] A semantic segmentation module, used for inputting the preprocessed remote sensing image into a preset semantic segmentation model to obtain a semantic image;

[0045] A detection module, used for inputting the semantic image into a preset target detection model to obtain information of a single container in the semantic image;

[0046] An output module is used to output port container layout information based on the information of each single container.

[0047] Wherein, the detection module comprises:

[0048] A frame selection unit, configured to input the semantic image into a preset detection model to frame a candidate area in the semantic image and obtain a rectangular target frame corresponding to the candidate area;

[0049] An input unit, used for inputting the semantic image into a convolutional neural network model to obtain a plurality of feature maps corresponding to the rectangular target box;

[0050] A fully connected unit, used for performing full connection on the multiple feature graphs to obtain the container location coordinates;

[0051] An angle calculation unit is used to calculate the tilt angle of the container based on the position coordinates of the container.

[0052] Wherein, the input unit comprises:

[0053] An acquisition subunit, configured to input the semantic image into a convolutional neural network model, and acquire a plurality of preset container features for the rectangular target frame in the convolutional neural network model and weight matrices corresponding to the plurality of preset container features respectively;

[0054] A convolution unit, configured to perform convolution processing on the semantic image based on the plurality of preset container features and the weight matrix to obtain a convolution processing result;

[0055] A pooling unit, used for performing pooling processing on the convolution processing result to obtain a pooling processing result;

[0056] The circulation unit is used to perform convolution and pooling alternating processing again for a corresponding number of times on the pooling processing result based on the preset number of alternating convolution and pooling times, so as to obtain multiple feature maps corresponding to the rectangular target box.

[0057] Wherein, the port container layout detection device also includes:

[0058] A first acquisition module, used for acquiring a preset first initial model and a preset first training data set, wherein the preset first training data includes a first training input image and a first theoretical output result;

[0059] An encoding module, configured to input the first training input image into the preset first initial model to encode the first training input image to obtain an encoding result;

[0060] A decoding module, used for decoding the encoding result to obtain a first actual output result;

[0061] A first comparison module, configured to compare the first actual output result with the first theoretical output result to obtain a first model loss value;

[0062] The first adjustment module is used to adjust and retrain the preset first initial model based on the first model loss value until the first model loss value is less than the preset first model threshold, so as to obtain the preset semantic segmentation model.

[0063] Wherein, the encoding module includes:

[0064] An input unit, configured to input the first training input image into a preset initial convolutional neural network to obtain high-level semantic information;

[0065] A recognition and classification unit, configured to classify and recognize the pixels in the first training input image based on the high-level semantic information to obtain a recognition and classification result;

[0066] An output unit is used to output a coding result based on the recognition and classification result.

[0067] Wherein, the port container layout detection device also includes:

[0068] A second acquisition module, used for acquiring a preset second initial model and a preset second training data set, wherein the preset second training data includes a second initial training image and a second theoretical output result;

[0069] A frame selection module, used for inputting the second initial training image into the preset second initial model to frame the candidate area in the second initial training image to obtain a rectangular target frame corresponding to the candidate area;

[0070] A convolution and pooling module is used to perform a preset number of convolution and pooling alternating processes on each of the second initial training images to obtain multiple feature maps corresponding to each of the rectangular target boxes.

[0071] A fully connected module, used for fully connecting the multiple feature maps corresponding to the rectangular target box to obtain a second actual output result;

[0072] A second comparison module, configured to compare the second actual output result with the second theoretical output result to obtain a second model loss value;

[0073] The second adjustment module is used to adjust and retrain the preset second initial model based on the second model loss value until the second model loss value is less than the preset second model threshold, so as to obtain the preset target detection model.

[0074] Wherein, the second acquisition module includes:

[0075] A format unification unit, configured to obtain a preset training remote sensing image, save the preset training remote sensing image as a preset text format image based on a preset data set standard, and read the preset text format image to obtain the second theoretical output result;

[0076] A standardization unit, used for performing cropping, data enhancement and normalization processing on the preset text format image to obtain a standard image;

[0077] The de-averaging unit is used to perform de-averaging processing on the standard image and convert the standard image into a preset data format to obtain the second initial training image.

[0078] The present invention also provides a port container layout detection device, which includes: a memory, a processor, and a program of the port container layout detection method stored in the memory and executable on the processor. When the program of the port container layout detection method is executed by the processor, the steps of the port container layout detection method as described above can be implemented.

[0079] The present invention also provides a readable storage medium, on which is stored a program for implementing the port container layout detection method. When the program of the port container layout detection method is executed by a processor, the steps of the port container layout detection method as described above are implemented.

[0080] The present application obtains remote sensing images, preprocesses the remote sensing images, and then inputs the preprocessed remote sensing images into a preset semantic segmentation model to obtain semantic images, and then inputs the semantic images into a preset target detection model to obtain the information of individual containers in the semantic images. Finally, based on the information of each individual container, the port container layout information is output. That is, the present application first obtains and preprocesses the remote sensing images, then extracts the semantic images through a preset semantic segmentation model, and then detects the semantic images through a preset detection model to obtain the information of each individual container in the semantic images. Finally, based on the information of each individual container, the port container layout information is output. That is, based on the preset semantic segmentation model and the preset detection model, the present application can accurately obtain the information of each container from the remote sensing images, and further obtain accurate and comprehensive port container layout information by mastering the information of each container, so as to achieve the purpose of fully mastering the port container layout, and then realize the high-efficiency dispatch of the port containers, and solve the technical problem of low dispatch efficiency of the port containers in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0082] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0083] Figure 1 It is a schematic diagram of the process of the first embodiment of the port container layout detection method of the present invention;

[0084] Figure 2 It is a flow chart of the second embodiment of the port container layout detection method of the present invention;

[0085] Figure 3 A schematic diagram of the process of the third embodiment of the port container layout detection method of the present invention

[0086] Figure 4 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention.

[0087] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0088] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0089] The present invention provides a port container layout detection method, which is applied to a port container layout detection device. In the first embodiment of the port container layout detection method of the present application, referring to Figure 1 , the port container layout detection method comprises:

[0090] Step S10, acquiring remote sensing images and preprocessing the remote sensing images;

[0091] In the present embodiment, it should be noted that the remote sensing image includes aerial photographs and satellite photographs, the remote sensing image includes a port area, and the port area is the main body of the remote sensing image, the remote sensing image is acquired, and the remote sensing image is preprocessed. Specifically, the remote sensing image is acquired by shooting with a preset aerial camera or a preset satellite, and then the remote sensing image is set to the same file format, and the remote sensing image is normalized to a uniform size, and the pixel value of the remote sensing image is within a preset pixel value range, wherein the preset pixel value range is a preset pixel value range that is pre-set. Further, the port area of ​​the remote sensing image is identified, and the port area is divided into a corresponding initialization category. Specifically, for example, the port area is distinguished from other areas in the remote sensing image by color, and then the image area with the same color as the port area is divided into an initialization category.

[0092] Step S20, inputting the preprocessed remote sensing image into a preset semantic segmentation model to obtain a semantic image;

[0093] In this embodiment, it should be noted that the preset semantic segmentation model is a trained convolutional neural network model. The preprocessed remote sensing image is input into the preset semantic segmentation model to obtain a semantic image. Specifically, the preprocessed remote sensing image is input into the preset semantic segmentation model to downsample the remote sensing image a preset number of times, wherein the downsampling method includes maximum pooling, average pooling, random pooling and summed area pooling, etc., so as to obtain high-level semantic information, wherein the high-level semantic information includes abstract features of the image subject, such as subject position, subject inclination and other information, and then by inputting the remote sensing image into the preset semantic segmentation model, the remote sensing image can be compressed into a series of latent variables, that is, the abstract features of the remote sensing image are extracted to obtain a feature map, but the image is discarded. The position information of the pixels is obtained, and these latent variables represent the high-level semantic information. Furthermore, the remote sensing image that has been downsampled is upsampled, wherein the upsampling method includes bilinear interpolation, deconvolution, depooling, etc., and then the spatial information of the image is gradually restored, that is, the position information of the pixels is gradually restored, so that the upsampled image is consistent with the resolution of the input remote sensing image, and a semantic image is obtained, wherein the semantic image is obtained by classifying the pixels in the remote sensing image according to the high-level semantic information, wherein the various areas include background areas, port areas, etc., and the purpose is to remove the correlation between the pixels in the image.

[0094] Step S30, inputting the semantic image into a preset target detection model to obtain information of a single container in the semantic image;

[0095] In this embodiment, it should be noted that the preset detection model is a preset trained remote sensing target detection model based on deep learning. The semantic image is input into the preset target detection model to obtain the single container information in the semantic image. Specifically, the semantic image is input into the preset detection model to identify the candidate area in the semantic image, wherein the candidate area refers to the area in the image that is most likely to be a container, and then the candidate area is matched with a preset target frame set, wherein the target frame set includes a preset number of target frames of a size, and then a selected target frame closest to the size of the candidate area is obtained, and the selected target frame is used to select the corresponding candidate area, wherein the candidate area is within the target frame, and then the semantic image is subjected to a preset number of convolution and pooling alternating processes, wherein the convolution refers to the process of multiplying the image matrix and the convolution kernel corresponding to the image element by element and then summing them to obtain the image feature value, wherein the convolution kernel refers to the weight matrix corresponding to the container feature, and the pooling refers to the convolution obtained by the convolution. The process of integrating image feature values ​​to obtain new feature values, further, obtaining shallow features through shallow convolution layers, wherein the shallow features include candidate area color, shape, etc., obtaining deep features through deep convolution layers, wherein the deep features include candidate area position coordinates, etc., and the candidate area inclination angle can be calculated according to the container position coordinates. For a feature of the candidate area, due to the alternating processing of convolution and pooling for a preset number of times, multiple feature maps will be obtained, and the multiple feature maps will be fully connected to output one-dimensional vectors corresponding to the multiple feature maps, wherein the one-dimensional vector includes a single candidate area information, and then based on the single candidate area information, it is determined whether the single candidate area information has a container. When the determination result is that it exists, the single container information is obtained, wherein the single container information includes container color, container category, container position coordinates, container inclination angle and other information.

[0096] The single container information includes the container location coordinates and the container tilt angle.

[0097] The step of inputting the semantic image into a preset target detection model to obtain the single container information in the semantic image comprises:

[0098] Step S31, inputting the semantic image into a preset detection model to select a candidate area in the semantic image and obtain a rectangular target box corresponding to the candidate area;

[0099] In this embodiment, the semantic image is input into a preset detection model to select a candidate area in the semantic image and obtain a rectangular target frame corresponding to the candidate area. Specifically, the semantic image is input into a preset detection model to identify the candidate area in the semantic image, and according to the size of the candidate area on the semantic image, wherein the candidate area refers to an area in the image that is highly likely to be a container, the candidate area is then matched with a preset target frame set, wherein the target frame set includes a preset number of target frames of a size, and then a selected target frame closest to the size of the candidate area is obtained. Specifically, the candidate area is selected with a target frame in the target frame set, and the area overlap ratio between the candidate area and the target frame is calculated, and the target frame with the largest area overlap ratio with the candidate area in the target frame set is the selected target frame, and the candidate area is inside the selected target frame.

[0100] Step S32, inputting the semantic image into a convolutional neural network model to obtain a plurality of feature maps corresponding to the rectangular target box;

[0101] In this embodiment, the semantic image is input into a convolutional neural network model to obtain a plurality of feature maps corresponding to the rectangular target box. Specifically, the semantic image is input into a convolutional neural network model, and the semantic image is divided into S*S grids. Feature extraction is performed on each of the grids based on the convolutional layer in the convolutional neural network model to obtain an initial feature map corresponding to each of the grids. Then, a pooling operation is performed on the initial feature map based on the pooling layer in the convolutional neural network model to reduce the dimension of the feature vector corresponding to the initial feature map. Furthermore, after performing a preset number of alternating convolution and pooling operations, a feature map corresponding to each of the grids is obtained. Furthermore, the feature maps corresponding to each of the grids are classified, and a plurality of feature maps corresponding to the same rectangular target box are extracted to obtain a plurality of feature maps corresponding to the rectangular target box.

[0102] The step of inputting the semantic image into a convolutional neural network model to obtain a plurality of feature maps corresponding to the rectangular target frame includes:

[0103] Step S321, inputting the semantic image into a convolutional neural network model, obtaining a plurality of preset container features for the rectangular target frame in the convolutional neural network model and weight matrices corresponding to the plurality of preset container features respectively;

[0104] In this embodiment, it should be noted that the convolution process can be understood as: the statistical characteristics of one part of the image feature are the same as those of other parts, that is, the statistical characteristics learned in this part can also appear in the corresponding other part, so the learned statistical characteristics are used as detectors and applied to any place of this image feature, that is, the statistical characteristics learned by the small-range image are convolved with the image features of the original large-size image. Mathematically, convolution can be the multiplication of the characteristic matrix of the corresponding image with multiple pre-detection matrices and the final summation to obtain the convolution processing result.

[0105] In this embodiment, the semantic image is input into a convolutional neural network model to obtain a plurality of preset container features for the rectangular target box in the convolutional neural network model and weight matrices corresponding to the plurality of preset container features, wherein the plurality of preset container features include features such as container color, container category, and container location coordinates. After obtaining the plurality of preset container features, statistical characteristics of the plurality of preset container features are obtained, and then the weight matrices corresponding to the plurality of preset container features are obtained.

[0106] Step S322, performing convolution processing on the semantic image based on the multiple preset container features and the weight matrix to obtain a convolution processing result;

[0107] In this embodiment, based on the multiple preset container features and the weight matrix, the semantic image is convolved to obtain a convolution result. Specifically, the image matrix corresponding to the semantic image is dot-multiplied by the weight matrix, and then the weights are summed to obtain the convolution result.

[0108] Step S323, performing pooling processing on the convolution processing result to obtain a pooling processing result;

[0109] In this embodiment, pooling processing includes maximum pooling, mean pooling and the like, and pooling processing is performed on the convolution processing result to obtain a pooling processing result. Specifically, the convolution processing result is first divided into a plurality of pixel matrices of preset sizes. If maximum pooling is used, the maximum pixel value of the pixel matrix is ​​used to replace the pixel matrix, thereby obtaining a new image matrix, that is, obtaining the pooling processing result.

[0110] Step S324: Based on a preset number of alternating convolution and pooling times, the pooling processing result is subjected to a corresponding number of alternating convolution and pooling processes again to obtain a plurality of feature maps corresponding to the rectangular target box.

[0111] In this embodiment, based on the preset number of alternating convolution and pooling times, the pooling processing result is again subjected to the corresponding number of convolution and pooling alternating processes to obtain multiple feature maps corresponding to the rectangular target box. Specifically, steps S322 to S323 are looped for a preset number of times to obtain feature maps corresponding to each of the rectangular target boxes, wherein when the preset number of times is small, the feature map represents the shallow features of the container, and when the preset number of times is large, the feature map represents the deep features of the container.

[0112] Step S33, performing full connection on the multiple feature graphs to obtain the container location coordinates;

[0113] In this embodiment, the multiple feature maps are fully connected to obtain the container location coordinates. Specifically, the feature maps of the rectangular target box are fully connected, wherein the full connection can be regarded as a special convolution process, and the result of the special convolution process is to obtain a one-dimensional vector corresponding to the image, that is, the multiple feature maps are converted into a one-dimensional vector through full connection, wherein one rectangular target box corresponds to a one-dimensional vector, and each one-dimensional vector includes the combination information of all features of the rectangular target box corresponding to the one-dimensional vector, wherein the combination information includes the candidate area determination result, the rectangular target box vertex coordinates are the container location coordinates, etc., wherein the candidate area determination result is obtained by analyzing the feature map, and it is determined whether there is a container in the candidate area, that is, by converting the multiple feature maps into a one-dimensional vector, it is determined whether there is a container in the candidate area, and when there is a container in the candidate area, the vertex coordinates of the rectangular target box are obtained.

[0114] Step S34, calculating the tilt angle of the container based on the container position coordinates.

[0115] In this embodiment, it should be noted that the container position coordinates refer to the coordinates of the vertices of the rectangular target frame corresponding to the container in the preset coordinate system in the image. Based on the container position coordinates, the container tilt angle is calculated. Specifically, assuming that the coordinates of the two vertices of the long side of the rectangular target frame are P1 (X1, Y1) and P2 (X2, Y2), the absolute spacing ΔW between P1 and P2 on the X axis is set to (X1-X2), and the absolute spacing ΔH on the Y axis is set to (Y1-Y2). Therefore, the container tilt angle θ can be calculated using the following formula.

[0116]

[0117] Step S40: outputting port container layout information based on the information of each single container.

[0118] In this embodiment, based on the information of each single container, the port container layout information is output. Specifically, based on the method of obtaining the single container information, the information of all containers in the port area is obtained, and the information of each single container is identified and stored in the same data set, wherein further, the data set is output, that is, the port container layout information is output, wherein the single container information of each container is identified in the data set, and the data set includes the number of containers and the distribution status of the containers in the image.

[0119] This embodiment obtains remote sensing images, preprocesses the remote sensing images, and then inputs the preprocessed remote sensing images into a preset semantic segmentation model to obtain semantic images, and then inputs the semantic images into a preset target detection model to obtain the information of individual containers in the semantic images. Finally, based on the information of each individual container, the port container layout information is output. That is, this embodiment first obtains and preprocesses the remote sensing images, then extracts the semantic images through a preset semantic segmentation model, and then detects the semantic images through a preset detection model to obtain the information of each individual container in the semantic images. Finally, based on the information of each individual container, the port container layout information is output. That is, based on the preset semantic segmentation model and the preset detection model, this embodiment can accurately obtain the information of each container from the remote sensing images, and further obtain accurate and comprehensive port container layout information by mastering the information of each container, so as to achieve the purpose of fully mastering the port container layout, and then achieve efficient scheduling of port containers, solving the technical problem of low efficiency of port container scheduling in the prior art.

[0120] Further, refer to Figure 2 Based on the first embodiment of the present application, in another embodiment of the port container layout detection method, the step of obtaining remote sensing images and preprocessing the remote sensing images includes:

[0121] Step A10, obtaining a preset first initial model and a preset first training data set, wherein the preset first training data includes a first training input image and a first theoretical output result;

[0122] In this embodiment, it should be noted that the preset first initial model is an untrained semantic segmentation model, wherein the preset first initial model convolution kernel is randomly determined, and the preset first training data set includes a first training input image and a first theoretical output result, wherein the first training input image is a remote sensing image that has undergone image preprocessing, wherein the image preprocessing includes unifying the image text format, unifying the image size, adjusting the image pixel value to be within a preset pixel value range, etc., and the first theoretical output result is the true label corresponding to the first training input image.

[0123] Step A20, inputting the first training input image into the preset first initial model to encode the first training input image to obtain an encoding result;

[0124] The step of encoding the first training input image to obtain an encoding result includes:

[0125] Step A21, inputting the first training input image into a preset initial convolutional neural network to obtain high-level semantic information;

[0126] In this embodiment, the first training input image is input into a preset initial convolutional neural network to obtain high-level semantic information. Specifically, the first training input image is input into a preset initial convolutional neural network to perform a preset number of convolution and pooling alternating processes on the first training input image to extract features of each pixel in the first training input image, that is, to obtain high-level semantic information.

[0127] Step A22, classifying and identifying the pixels in the first training input image based on the high-level semantic information to obtain a recognition and classification result;

[0128] In this embodiment, based on the high-level semantic information, the pixels in the first training input image are classified and identified to obtain identification and classification results. Specifically, the features of each pixel are identified, and the pixel points are classified, and the pixels of the same category are divided into a pixel set. For example, assuming that the first training input image includes a port area and a background area, after the first training input image is input into a preset initial convolutional neural network, the extracted pixel features are identified, and the probabilities P1 and P2 that the pixel points belong to the port area and the background area, respectively, are calculated, and P1+P2=1. If P1 is greater than P2, the pixel point belongs to the port area. If P1 is less than P2, the pixel point belongs to the background area, and then the pixel points are divided into two categories, one corresponding to the port area and the other corresponding to the background area.

[0129] Step A23: output the encoding result based on the recognition and classification result.

[0130] In this embodiment, it should be noted that the recognition and classification result is the result of downsampling the first training input image, that is, each pixel in the first training input image is recognized and classified based on the high-level semantic information obtained by downsampling, and then based on the recognition and classification result, an encoding result is output, wherein the encoding result includes the recognition and classification results of the pixel points. Specifically, the encoding result is the image matrix output by the preset convolutional neural network, wherein the pixel values ​​in the image matrix represent the recognition and classification results of the pixel points. For example, assuming that the pixel values ​​in the image matrix are composed of 0 and 1, a pixel value of 1 indicates that the corresponding pixel point belongs to the port area, and a pixel value of 0 indicates that the corresponding pixel point belongs to the background area.

[0131] Step A30, decoding the encoding result to obtain a first actual output result;

[0132] In this embodiment, the first actual output result is an actual output semantic image, and the resolution of the actual output semantic image should be consistent with that of the first training input image. The encoding result is decoded to obtain the first actual output result. Specifically, the image matrix corresponding to the encoding result is deconvolved, that is, the image matrix corresponding to the encoding result is multiplied by the transposed weight matrix to obtain a semantic image matrix corresponding to the actual output semantic image. The actual output semantic image corresponding to the semantic image matrix is ​​output to obtain the first actual output result. According to the classification of the pixel points, different colors are used to distinguish the port area from the background area in the actual output semantic image.

[0133] Step A40, comparing the first actual output result with the first theoretical output result to obtain a first model loss value;

[0134] In this embodiment, the first actual output result is compared with the first theoretical output result to obtain a first model loss value, wherein the first theoretical output result is a true label corresponding to the first training input image. Specifically, assuming that the true label is the shape and area of ​​the port area in the first training input image, the port area in the actual output semantic image and the port area in the first training input image are overlapped to obtain the ratio of the non-overlapping area of ​​the port area to the area of ​​the port area in the first training input image, that is, to obtain the first model loss value.

[0135] Step A50, based on the first model loss value, adjust and retrain the preset first initial model until the first model loss value is less than the preset first model threshold, so as to obtain the preset semantic segmentation model.

[0136] In this embodiment, based on the first model loss value, the preset first initial model is adjusted and retrained until the first model loss value is less than the preset first model threshold, so as to obtain the preset semantic segmentation model. Specifically, the first model loss value is compared with the preset first model threshold. When the first model loss value is less than the preset first model threshold, the preset first initial model is set as the preset semantic segmentation model. When the first model loss value is greater than or equal to the preset first model threshold, based on the first model loss value, the convolution kernel in the preset convolutional neural network in the preset first initial model is adjusted, that is, the weight matrix is ​​adjusted, and then the preset first initial model is retrained to achieve the effect of reducing the first model loss value until the first model loss value is less than the preset first model threshold.

[0137] This embodiment obtains a preset first initial model and a preset first training data set, wherein the preset first training data includes a first training input image and a first theoretical output result, and then inputs the first training input image into the preset first initial model to encode the first training input image to obtain an encoding result, and then decodes the encoding result to obtain a first actual output result, and further, compares the first actual output result with the first theoretical output result to obtain a first model loss value, and finally adjusts and retrains the preset first initial model based on the first model loss value until the first model loss value is less than the preset first model threshold, thereby obtaining the preset semantic segmentation model. That is, this embodiment first obtains the preset first initial model and the preset first training data set, and then obtains the preset semantic segmentation model by training the preset first initial model, thereby providing a basis for realizing the conversion of the remote sensing image into a semantic image, and laying a foundation for solving the technical problem of low efficiency of port container scheduling in the prior art.

[0138] Further, refer to Figure 3 Based on the first embodiment and the second embodiment of the present application, in another embodiment of the port container layout detection method, the step of obtaining remote sensing images and preprocessing the remote sensing images includes:

[0139] Step B10, obtaining a preset second initial model and a preset second training data set, wherein the preset second training data includes a second initial training image and a second theoretical output result;

[0140] In this embodiment, it should be noted that the preset second initial model is an untrained target detection model, that is, the weight matrix in the preset second initial model is randomly determined, and the preset second training data includes a second initial training image and a second theoretical output result, the second initial training image is a semantic image obtained by image processing of a remote sensing image, and the second theoretical output result is the actual port container layout information corresponding to the second initial training image.

[0141] Wherein, in step B10, the step of obtaining a preset second training data set includes:

[0142] Step B11, obtaining a preset training remote sensing image, and based on a preset data set standard, saving the preset training remote sensing image as a preset text format image, and reading the preset text format image to obtain the second theoretical output result;

[0143] In this embodiment, a preset training remote sensing image is obtained, and based on a preset data set standard, the preset training remote sensing image is saved as a preset text format image, and the preset text format image is read to obtain the second theoretical output result. Specifically, the preset training remote sensing image is extracted from a preset sample database, and based on a preset data set standard, wherein the data set standard includes a VOC standard, the preset training remote sensing image is saved as a preset text format image, for example, saved as an xml file. Further, the xml file is read to obtain the actual port container layout information in the image, including container location coordinates, container quantity, container category, container inclination angle and other information, that is, to obtain the second theoretical output result.

[0144] Step B12, performing cropping, data enhancement and normalization processing on the preset text format image to obtain a standard image;

[0145] In this embodiment, the preset text format image is cropped, data enhanced and normalized to obtain a standard image, wherein the normalization processing refers to performing data normalization processing on the data to standardize the data, wherein data standardization includes that the image size corresponding to the image data is consistent, the data conforms to the standard normal distribution, and the data is within a fixed range, etc. Specifically, the preset text format image is cropped to obtain a port area picture in the preset text format image, and data enhancement is performed on the port area picture, including random perturbation of image color values, Gaussian blur, image sharpening, image flipping, translation and other operations, and the enhanced image is size normalized to obtain a fixed-size picture, that is, a standard image is obtained.

[0146] Step B13, performing a mean removal process on the standard image, and converting the standard image into a preset data format to obtain the second initial training image.

[0147] In this embodiment, the standard image is de-meaned and converted into a preset data format to obtain the second initial training image. Specifically, based on the stationarity of the image, that is, the statistical characteristics of any part of the image should be the same as other parts, the standard image is zero-meaned, that is, the average brightness value of the image is removed. Furthermore, in order to facilitate model training, the standard image is converted into a tfrecord data format to obtain the second initial training image.

[0148] Step B20, inputting the second initial training image into the preset second initial model to select the candidate area in the second initial training image to obtain a rectangular target box corresponding to the candidate area;

[0149] In this embodiment, the second initial training image is input into the preset second initial model to select the candidate area in the second initial training image and obtain a rectangular target frame corresponding to the candidate area. Specifically, the second initial training image is input into the preset second initial model, the candidate area in the second initial training image is identified, and a suitable selected target frame is matched according to the size of the candidate area, and the size of the rectangular target frame is larger than the size of the candidate area. More specifically, the candidate area is selected by the target frame in the target frame set, and the area overlap rate between the candidate area and the target frame is calculated. The target frame with the largest area overlap rate with the candidate area in the target frame set is the selected target frame, and the candidate area is inside the selected target frame.

[0150] Step B30, performing a preset number of convolution and pooling alternating processes on each of the second initial training images to obtain a plurality of feature maps corresponding to each of the rectangular target boxes;

[0151] In this embodiment, each of the second initial training images is subjected to a preset number of convolution and pooling alternating processes to obtain a plurality of feature maps corresponding to each of the rectangular target frames. Specifically, the second initial training image is divided into an S*S grid, that is, the second initial training image includes S*S image grids, and then the second initial training image is input into the convolutional neural network in the preset second initial model, and each of the second initial training images is subjected to a preset number of convolution and pooling alternating processes, that is, each image grid is subjected to a preset number of convolution and pooling alternating processes, and features of each image grid are extracted to obtain a plurality of grid feature maps. Furthermore, the grid feature maps belonging to the same rectangular target frame are extracted to obtain a plurality of feature maps corresponding to each of the rectangular target frames.

[0152] Step B40, performing full connection on multiple feature maps corresponding to the rectangular target box to obtain a second actual output result;

[0153] In this embodiment, the multiple feature maps corresponding to the rectangular target box are fully connected to obtain the second actual output result. Specifically, the multiple feature maps corresponding to a single rectangular target box are fully connected to obtain a one-dimensional vector corresponding to the rectangular target box, wherein the one-dimensional vector includes all feature information of the rectangular target box. Further, the multiple feature maps corresponding to all rectangular target boxes are fully connected respectively to obtain feature information of all target boxes, wherein the feature information includes target box vertex coordinates, target box quantity, target box type, and target box inclination angle. That is, the theoretical port container layout information is obtained, that is, the second actual output result is obtained.

[0154] Step B50, comparing the second actual output result with the second theoretical output result to obtain a second model loss value;

[0155] In this embodiment, the second actual output result refers to the actual port container layout information, and the second theoretical output result refers to the theoretical port container layout information. The second actual output result is compared with the second theoretical output result to obtain a second model loss value. Specifically, the actual port container layout information is compared with the theoretical port container layout information to calculate the comprehensive information error, wherein the comprehensive error includes container position error, container quantity error, container tilt angle error and container category error, etc., that is, the second model loss value is obtained.

[0156] Step B60, based on the second model loss value, adjust and retrain the preset second initial model until the second model loss value is less than the preset second model threshold, thereby obtaining the preset target detection model.

[0157] In this embodiment, based on the second model loss value, the preset second initial model is adjusted and retrained until the second model loss value is less than the preset second model threshold, so as to obtain the preset target detection model. Specifically, the second model loss value is compared with the preset second model threshold. When the second model loss value is less than the preset second model threshold, the preset second initial model is set as the preset semantic segmentation model. When the second model loss value is greater than or equal to the preset second threshold, based on the second model loss value, the convolution kernel in the preset convolutional neural network in the preset second initial model is adjusted, that is, the weight matrix is ​​adjusted, and then the preset second initial model is retrained to achieve the effect of reducing the second model loss value until the second model loss value is less than the preset second model threshold.

[0158] This embodiment obtains a preset second initial model and a preset second training data set, wherein the preset second training data includes a second initial training image and a second theoretical output result, and then inputs the second initial training image into the preset second initial model to select the candidate area in the second initial training image to obtain a rectangular target box corresponding to the candidate area, and then performs a preset number of convolution and pooling alternating processes on each of the second initial training images to obtain multiple feature maps corresponding to each of the rectangular target boxes, and further, fully connects the multiple feature maps corresponding to the rectangular target boxes to obtain a second actual output result, and then compares the second actual output result with the second theoretical output result to obtain a second model loss value, and finally, based on the second model loss value, adjusts and retrains the preset second initial model until the second model loss value is less than the preset second model threshold to obtain the preset target detection model. That is, this embodiment, based on the acquisition of the preset second initial model and the preset second training data set, trains the preset second initial model, and then obtains the preset detection model, and then based on the preset detection model, it can realize the acquisition of the port container layout information in the semantic image. Therefore, this embodiment lays a foundation for solving the technical problem of low efficiency of port container scheduling in the prior art.

[0159] Reference Figure 4 , Figure 4 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention.

[0160] like Figure 4As shown, the port container layout detection device may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the processor 1001 and the memory 1005. The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0161] Optionally, the port container layout detection device may also include a rectangular user interface, a network interface, a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, etc. The rectangular user interface may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional rectangular user interface may also include a standard wired interface and a wireless interface. The network interface may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0162] Those skilled in the art will understand that Figure 3 The structure of the port container layout detection equipment shown in the figure does not constitute a limitation on the port container layout detection equipment, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0163] like Figure 4 As shown, the memory 1005 as a computer storage readable storage medium may include an operating system, a network communication module and a port container layout detection program. The operating system is a program that manages and controls the hardware and software resources of the port container layout detection device, and supports the operation of the port container layout detection program and other software and / or programs. The network communication module is used to realize the communication between the components inside the memory 1005, and the communication with other hardware and software in the port container layout detection system.

[0164] exist Figure 4 In the port container layout detection device shown, the processor 1001 is used to execute the port container layout detection program stored in the memory 1005 to implement the steps of any of the above-mentioned port container layout detection methods.

[0165] The specific implementation of the port container layout detection device of the present invention is basically the same as the various embodiments of the port container layout detection method described above, and will not be repeated here.

[0166] The present invention also provides a port container layout detection device, the port container layout detection device comprising:

[0167] A preprocessing module, used for acquiring remote sensing images and preprocessing the remote sensing images;

[0168] A semantic segmentation module, used for inputting the preprocessed remote sensing image into a preset semantic segmentation model to obtain a semantic image;

[0169] A detection module, used for inputting the semantic image into a preset target detection model to obtain information of a single container in the semantic image;

[0170] An output module is used to output port container layout information based on the information of each single container.

[0171] Wherein, the detection module comprises:

[0172] A frame selection unit, configured to input the semantic image into a preset detection model to frame a candidate area in the semantic image and obtain a rectangular target frame corresponding to the candidate area;

[0173] An input unit, used for inputting the semantic image into a convolutional neural network model to obtain a plurality of feature maps corresponding to the rectangular target box;

[0174] A fully connected unit, used for performing full connection on the multiple feature graphs to obtain the container location coordinates;

[0175] An angle calculation unit is used to calculate the tilt angle of the container based on the position coordinates of the container.

[0176] Wherein, the input unit comprises:

[0177] An acquisition subunit, configured to input the semantic image into a convolutional neural network model, and acquire a plurality of preset container features for the rectangular target frame in the convolutional neural network model and weight matrices corresponding to the plurality of preset container features respectively;

[0178] A convolution unit, configured to perform convolution processing on the semantic image based on the plurality of preset container features and the weight matrix to obtain a convolution processing result;

[0179] A pooling unit, used for performing pooling processing on the convolution processing result to obtain a pooling processing result;

[0180] The circulation unit is used to perform convolution and pooling alternating processing again for a corresponding number of times on the pooling processing result based on the preset number of alternating convolution and pooling times, so as to obtain multiple feature maps corresponding to the rectangular target box.

[0181] Wherein, the port container layout detection device also includes:

[0182] A first acquisition module, used for acquiring a preset first initial model and a preset first training data set, wherein the preset first training data includes a first training input image and a first theoretical output result;

[0183] An encoding module, configured to input the first training input image into the preset first initial model to encode the first training input image to obtain an encoding result;

[0184] A decoding module, used for decoding the encoding result to obtain a first actual output result;

[0185] A first comparison module, configured to compare the first actual output result with the first theoretical output result to obtain a first model loss value;

[0186] The first adjustment module is used to adjust and retrain the preset first initial model based on the first model loss value until the first model loss value is less than the preset first model threshold, so as to obtain the preset semantic segmentation model.

[0187] Wherein, the encoding module includes:

[0188] An input unit, configured to input the first training input image into a preset initial convolutional neural network to obtain high-level semantic information;

[0189] A recognition and classification unit, configured to classify and recognize the pixels in the first training input image based on the high-level semantic information to obtain a recognition and classification result;

[0190] An output unit is used to output a coding result based on the recognition and classification result.

[0191] Wherein, the port container layout detection device also includes:

[0192] A second acquisition module, used for acquiring a preset second initial model and a preset second training data set, wherein the preset second training data includes a second initial training image and a second theoretical output result;

[0193] A frame selection module, used for inputting the second initial training image into the preset second initial model to frame the candidate area in the second initial training image to obtain a rectangular target frame corresponding to the candidate area;

[0194] A convolution and pooling module is used to perform a preset number of convolution and pooling alternating processes on each of the second initial training images to obtain multiple feature maps corresponding to each of the rectangular target boxes.

[0195] A fully connected module, used for fully connecting the multiple feature maps corresponding to the rectangular target box to obtain a second actual output result;

[0196] A second comparison module, configured to compare the second actual output result with the second theoretical output result to obtain a second model loss value;

[0197] The second adjustment module is used to adjust and retrain the preset second initial model based on the second model loss value until the second model loss value is less than the preset second model threshold, so as to obtain the preset target detection model.

[0198] Wherein, the second acquisition module includes:

[0199] A format unification unit, configured to obtain a preset training remote sensing image, save the preset training remote sensing image as a preset text format image based on a preset data set standard, and read the preset text format image to obtain the second theoretical output result;

[0200] A standardization unit, used for performing cropping, data enhancement and normalization processing on the preset text format image to obtain a standard image;

[0201] The de-averaging unit is used to perform de-averaging processing on the standard image and convert the standard image into a preset data format to obtain the second initial training image.

[0202] The specific implementation of the port container layout detection device of the present invention is basically the same as the various embodiments of the port container layout detection method described above, and will not be repeated here.

[0203] The present invention provides a readable storage medium, which stores one or more programs. The one or more programs can also be executed by one or more processors to implement the steps of any of the above-mentioned port container layout detection methods.

[0204] The specific implementation of the readable storage medium of the present invention is basically the same as the above-mentioned embodiments of the port container layout detection method, and will not be repeated here.

[0205] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent scope of the present invention.

Claims

1. A method for detecting port container layout, characterized in that: The port container layout detection method comprises: Acquire a preset first initial model and a preset first training data set, wherein the preset first training data set includes a first training input image and a first theoretical output result; Inputting the first training input image into the preset first initial model to encode the first training input image to obtain an encoding result; The step of encoding the first training input image to obtain an encoding result includes: Inputting the first training input image into a preset initial convolutional neural network to obtain high-level semantic information; Based on the high-level semantic information, classify and identify the pixels in the first training input image to obtain a recognition and classification result; Based on the identification and classification results, output a coding result; Decoding the encoding result to obtain a first actual output result; Comparing the first actual output result with the first theoretical output result to obtain a first model loss value; Based on the first model loss value, adjusting and retraining the preset first initial model until the first model loss value is less than the preset first model threshold, thereby obtaining a preset semantic segmentation model; Acquire remote sensing images and preprocess the remote sensing images; Inputting the preprocessed remote sensing image into a preset semantic segmentation model to obtain a semantic image; Inputting the semantic image into a preset target detection model to obtain information about a single container in the semantic image; The single container information includes the container location coordinates and the container tilt angle. The step of inputting the semantic image into a preset target detection model to obtain the single container information in the semantic image comprises: Inputting the semantic image into a preset detection model to select a candidate area in the semantic image and obtain a rectangular target box corresponding to the candidate area; Inputting the semantic image into a convolutional neural network model to obtain a plurality of feature maps corresponding to the rectangular target box; Performing full connection on the multiple feature graphs to obtain the location coordinates of the container; Calculating the tilt angle of the container based on the container position coordinates; Based on the individual container information, port container layout information is output.

2. The method for detecting the layout of a port container according to claim 1, characterized in that: The step of inputting the semantic image into a convolutional neural network model to obtain a plurality of feature maps corresponding to the rectangular target frame comprises: Inputting the semantic image into a convolutional neural network model, obtaining a plurality of preset container features for the rectangular target frame in the convolutional neural network model and weight matrices corresponding to the plurality of preset container features respectively; Based on the multiple preset container features and the weight matrix, performing convolution processing on the semantic image to obtain a convolution processing result; Performing pooling processing on the convolution processing result to obtain a pooling processing result; Based on a preset number of alternating convolution and pooling times, the pooling processing result is again subjected to a corresponding number of alternating convolution and pooling processes to obtain a plurality of feature maps corresponding to the rectangular target box.

3. The method for detecting the layout of a port container according to claim 1, characterized in that: The step of obtaining remote sensing images and preprocessing the remote sensing images includes: Acquire a preset second initial model and a preset second training data set, wherein the preset second training data includes a second initial training image and a second theoretical output result; Inputting the second initial training image into the preset second initial model to select a candidate area in the second initial training image and obtain a rectangular target box corresponding to the candidate area; Performing a preset number of convolution and pooling alternating processes on each of the second initial training images to obtain a plurality of feature maps corresponding to each of the rectangular target frames; Fully connect multiple feature maps corresponding to the rectangular target box to obtain a second actual output result; Comparing the second actual output result with the second theoretical output result to obtain a second model loss value; Based on the second model loss value, the preset second initial model is adjusted and retrained until the second model loss value is less than the preset second model threshold, so as to obtain the preset target detection model.

4. The method for detecting the layout of a port container according to claim 3, characterized in that: The step of obtaining a preset second training data set includes: Acquire a preset training remote sensing image, and based on a preset data set standard, save the preset training remote sensing image as a preset text format image, and read the preset text format image to obtain the second theoretical output result; Performing cropping, data enhancement and normalization processing on the preset text format image to obtain a standard image; The standard image is subjected to a mean removal process and converted into a preset data format to obtain the second initial training image.

5. A port container layout detection device, characterized in that: The port container layout detection device comprises: a memory, a processor and a program stored in the memory for implementing the port container layout detection method. The memory is used to store a program for implementing a port container layout detection method; The processor is used to execute a program for implementing the port container layout detection method, so as to implement the steps of the port container layout detection method according to any one of claims 1 to 4.

6. A port container layout detection device, characterized in that: The port container layout detection device is applied to the port container layout detection equipment according to claim 5, and the port container layout detection device comprises: A preprocessing module, used for acquiring remote sensing images and preprocessing the remote sensing images; A semantic segmentation module, used for inputting the preprocessed remote sensing image into a preset semantic segmentation model to obtain a semantic image; A detection module, used for inputting the semantic image into a preset target detection model to obtain information of a single container in the semantic image; An output module, used for outputting port container layout information based on each of the single container information; Wherein, the port container layout detection device comprises: Acquire a preset first initial model and a preset first training data set, wherein the preset first training data set includes a first training input image and a first theoretical output result; Inputting the first training input image into the preset first initial model to encode the first training input image to obtain an encoding result; Decoding the encoding result to obtain a first actual output result; Comparing the first actual output result with the first theoretical output result to obtain a first model loss value; Based on the first model loss value, the preset first initial model is adjusted and retrained until the first model loss value is less than the preset first model threshold, so as to obtain the preset semantic segmentation model.

7. A readable storage medium, characterized in that: The readable storage medium stores a program for implementing the port container layout detection method, and the program for implementing the port container layout detection method is executed by a processor to implement the steps of the port container layout detection method according to any one of claims 1 to 4.

Images