# US 2019/0171879 A1 · Method of extracting warehouse in port from hierarchically screened remote sensing image

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2019/0171879 A1
- Document ID: US20190171879A1
- Status: Published application
- Kind: A1
- Application number: 16/205,350
- Filed: 2018-11-30
- Published: 2019-06-06
- Earliest claimed: 2017-12-04
- Assignees: Transport Planning and Research Institute Ministry of Transport
- Applicants: Transport Planning and Research Institute Ministry of Transport
- Inventors: Yue QI; Xiangjun NIE; Min DONG; Dachuan WANG; Lu SUN; Jia TIAN; Zhuo FANG; Rui LI; Lei MEI; Jing YANG
- Web page: https://patentgrep.com/patent/US20190171879A1

## CPC

- G06T 7/12: Edge-based segmentation
- G06F 18/2135: based on approximation criteria, e.g. principal component analysis
- G06T 7/13: Edge detection
- G06T 7/149: involving deformable models, e.g. active contour models
- G06T 7/40: Analysis of texture (depth or shape recovery from texture G06T7/529)
- G06V 10/464: {using a plurality of salient features, e.g. bag-of-words [BoW] representations}
- G06V 10/7715: {Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods}
- G06V 20/176: {Urban or other man-made structures}
- G06T 2207/10032: Satellite or aerial image; Remote sensing
- G06T 2207/20016: Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
- G06T 2207/20024: Filtering details
- G06T 2207/20116: Active contour; Active surface; Snakes
- G06T 2207/30184: Infrastructure
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

A method of extracting a warehouse in a port from a hierarchically screened remote sensing image includes the following steps: first, recognizing a texture feature of a remote sensing image and extracting edge lines of a coast of a port; then, selecting a sample of an optional irregular texture region and forming, through a CA transformation, principal component images of different hierarchies by taking a ratio of a between-class difference to an intra-class difference being maximum as an optimization condition; sequentially, extracting a correlation relationship of the warehouse in the port, and forming a feature point set with recognized warehouses to be analyzed; and last, extracting a feature of a visually sensitive image through a scene image to obtain a feedback selection of a real scene image to extract the warehouse in the port accurately.

## Claims

### Claim 1

A method of extracting a warehouse in a port from a hierarchically screened remote sensing image, comprising the following steps:
S100: extracting a coastline of the port based on an active contour model, successively performing a texture feature recognition on any region in a remote sensing image to form a sea area texture region and an irregular texture region, and extracting edge lines of the coastline of the port;
S200: extracting principal component images of a plurality of hierarchies using insufficient spectrum features, selecting a sample of the irregular texture region, and forming, through a CA transformation, principal component images of different hierarchies with different difference values by taking a ratio of a between-class difference to an intra-class difference being maximum as an optimization condition;
S300: accurately recognizing the warehouse in the port using a spatial relationship feature, extracting a correlation relationship of the warehouse in the port from the principle component images, and forming a feature point set with recognized warehouses to be analyzed; and
S400: extracting a feature of a visually sensitive image from the feature point set through a scene image based on WTA visual rapid adaptation selection to obtain a feedback selection of a real scene image to extract the warehouse in the port.

### Claim 2 (depends on 1)

The method of extracting the warehouse in the port from the hierarchically screened remote sensing image according to claim 1, wherein
in step S100, a gray level co-occurrence matrix method is used for the texture feature recognition, and the texture feature recognition comprises the following steps:
S101: selecting a region of the remote sensing image, and setting that the region has L gray level values, and a gray level co-occurrence matrix corresponding to the region is a matrix having L×L orders;
S102: selecting a position (i, j) in the matrix, where (i, j=1, 2, . . . , L), wherein, an element at the optional position is a first pixel at a fixed distance from a second pixel having a gray level of i and a gray level of j, wherein the following fixed positional relationship exists between the first pixel and the second pixel: ζ=(DX, DY), where ζ is a displacement, and DX and DY are distances in two directions;
S103: extracting, according to a positional relationship between gray level co-occurrence matrixes, a texture feature quantity including an Angular Second Moment (ASM) and a contrast CON, wherein ASM=Σi=1LΣj=1LP2 (i, j), and CON=Σn=1Ln2[Σi=1LΣj=1LP(i, j)]|i−j|, where P is a feature vector at the position (i,j), and n is the number of times the extraction is performed.

### Claim 3 (depends on 1)

The method of extracting the warehouse in the ort from the hierarchically screened remote sensing image according to claim 1, wherein
in step S100, extracting edge lines of the coast of the port between the sea area texture region and the irregular texture region using the filter algorithm and optimizing the edge lines of the coast of the port using the filter algorithm comprises the following steps: first, acquiring discrete data of a texture feature and selecting a lowest center frequency when extracting an image feature in a filter using a discretized Gabor template matrix and an image data matrix convolution, and then carrying out a frequency spectrum superposition calculation again to obtain a filtered image.

### Claim 4 (depends on 1)

The method of extracting the warehouse in the port from the hierarchically screened remote sensing image according to claim 1, wherein
in step S200, before the CA transformation is executed, a maximized ratio of a between-class variance of an optional data set to an intra-class variance of the optional data set is extracted according to the following linear transformation formula: Y=TX, where T is an ideal transformation matrix, so as to ensure the maximum separability of the optional data set to provide optimized basic data for the CA transformation

### Claim 5 (depends on 4)

The method of extracting the warehouse in the port from the hierarchically screened remote sensing image according to claim 4, wherein
a specific algorithm of the ideal transformation matrix is as follows:
S201: σA is set as a standard deviation of a class 1 and a class 2 obtained after the CA transformation, σw1 and σw2 are set as intra-class standard deviations of the class 1 and the class 2, and σw is set as an average value of σw1 and σw2;
S202: a relationship between a transformed variance and an untransformed variance is as follows:
σw2=tTSwt, σA2=tTSAt, where Sw and SA are an intra-class scatter matrix and a between-class scatter matrix of a given sample, and t is a mapping transformation vector; and
S203: the mapping transformation vector t is set as a special value of a ratio σA2/aσW2, of the between-class variance and the intra-class variance, where, λ=σA2/σW2=tTSAt/tTSwt, when the mapping transformation vector t approximates a maximum value, (SA−ΛSW) T=0, where Λ represents a diagonal matrix consisting of all feature values λ, and a matrix T composed of all column vectors t is a desired ideal transformation matrix.

### Claim 6 (depends on 1)

The method of extracting the warehouse in the port from the hierarchically screened remote sensing image according to claim 1, wherein
in step S300, a correlation relationship of the warehouse in the port is extracted, the correlation relationship includes a point feature, a line feature and a plane feature included in spatial features; a hierarchical relationship feature of the correlation relationship is acquired by extracting a hierarchy attribute of the remote sensing image based on a spectral feature of the remote sensing image.

### Claim 7 (depends on 6)

The method of extracting the warehouse in the port from the hierarchically screened remote sensing image according to claim 6, wherein
the correlation relationship of the warehouse in the port includes a road relationship, a transshipment square relationship and an enclosure relationship of the warehouse in the port, and attributes of a whole are extracted using a spatial correlation relationship of the warehouse in the port.

### Claim 8 (depends on 1)

The method of extracting the warehouse in the port from the hierarchically screened remote sensing image according to claim 1, wherein
in step S400, the visually sensitive image includes a gray level, colors, edges, textures and a motion, a visual saliency map of each position in a scene image is obtained according to synthesized features, and a mutual competition of a plurality of the visual saliency maps transfers an inhibition of return of focus.

### Claim 9 (depends on 8)

The method of extracting the warehouse in the port from the hierarchically screened remote sensing image according to claim 8, wherein
the mutual competition and the inhibition of the visual saliency map comprises the following steps:
S401: selecting a plurality of parallel and separable feature maps from the feature point set, and recording a hierarchy attribute of each position in a feature dimension on a feature map to obtain a saliency of each position in different feature dimensions;
S402: merging saliencies of different feature maps to obtain a total saliency measure, and guiding a visual attention process; and
S403: dynamically selecting, through a WTA network, a position with the highest saliency from the saliency map as a Focus Of Attention (FOA), and then performing the processing circularly through the inhibition of return until a real scene image is obtained.

### Claim 10 (depends on 1)

The method of extracting the warehouse in the port from the hierarchically screened remote sensing image according to claim 1, further comprising:
a step S500 of tracking a nonlinear filtering feature, comprising: separately extracting a filtering feature obtained in step S100 using hierarchical image attributes extracted through S100, S200, S300 and S400, and performing a tracking in a remote sensing analysis image according to a texture feature extracted according to a hierarchical analysis to compensate for an attribute that cannot be directly extracted by tracking the texture feature to form an interpreted remote sensing image, and comparing the interpreted remote sensing image after being formed with a real scene image in step S400 to remove an inaccurate tracked texture and keep a rational tracked texture.