CN114863258A - Method for detecting small target based on visual angle conversion in sea-sky-line scene - Google Patents
Method for detecting small target based on visual angle conversion in sea-sky-line scene Download PDFInfo
- Publication number
- CN114863258A CN114863258A CN202210786036.1A CN202210786036A CN114863258A CN 114863258 A CN114863258 A CN 114863258A CN 202210786036 A CN202210786036 A CN 202210786036A CN 114863258 A CN114863258 A CN 114863258A
- Authority
- CN
- China
- Prior art keywords
- sea
- coordinate information
- image
- sky
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/778—Active pattern-learning, e.g. online learning of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
- Radar Systems Or Details Thereof (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for detecting a small target based on visual angle conversion in a sea-sky-line scene, which comprises the following steps: acquiring an image to be detected; identifying a sea-sky-line; framing an effective rectangular area of an image to be detected; dividing the effective rectangular area into N 2 Each image block, wherein an overlapping area exists between every two adjacent image blocks; will N 2 Arranging the image blocks according to N rows and N columns to obtain a recombined image; detecting sea surface targets in the recombined image by using a deep learning network model, obtaining first coordinate information of each sea surface target, and combining the first coordinate information of each sea surface target into a first coordinate information set; respectively converting the first coordinate information of each sea surface target into the first coordinate information of the sea surface target to be detectedSecond coordinate information in the image, and forming a second set of coordinate information. The invention improves the detection precision of small target ships and small target buoys near the sea-sky-line through the selection of the effective area and the recombination of the image blocks.
Description
Technical Field
The invention belongs to the technical field of target identification, and particularly relates to a method for detecting a small target based on visual angle conversion in a sea-sky-line scene.
Background
At present, the target detection technology plays more and more important roles in various fields and is gradually mature. The existing mature target detection technology is mostly based on a deep learning method. In general, a deep learning method scales input image data to a fixed size in an image preprocessing stage, and then detects a target in the scaled image data by loading a model, for example, applying a relatively wide YOLO algorithm model and SSD algorithm model.
In the field of sea surface target detection, small target ships, small target buoys and the like near sea antennas cause targets to be extremely small due to the long distance. At present, when a small target ship and a small target buoy near a sea antenna are detected, the following problems exist during the operation of a traditional deep learning network model:
1. in a long-distance sea-sky-line image input into the deep learning network model, the long-distance sea-sky-line image is a higher-resolution image, such as 5472 × 3648, however, the proportion of pixels of a small target near a sea-sky-line in the whole image is small, most of the pixels are regions such as ocean wave sky and the like which are not concerned by a user, and therefore most of operation time is applied to unrelated regions when a traditional algorithm is operated; 2. in an image preprocessing stage of a deep learning network model, high-resolution image data is compressed to a resolution of 640 × 640 pixels, even in order to compress the computational efficiency to a resolution of 416 × 416 pixels, small targets near a sea antenna become smaller, and meanwhile, more interference pixel points are introduced, so that detection failure is caused; 3. under a complex sea-sky-line scene, airplanes, flying birds and the like in the sky near the sea-sky-line cause large interference to the identification of small target ships, small target buoys and the like near the sea-sky-line, and the interference cannot be eliminated.
Disclosure of Invention
The invention aims to overcome one or more defects in the prior art and provides a method for detecting a small target based on view angle conversion in a sea-sky-line scene.
The purpose of the invention is realized by the following technical scheme:
the method for detecting the small target based on the view angle conversion in the sea-sky-line scene specifically comprises the following steps:
acquiring an image to be detected;
identifying a sea-sky line in an image to be detected;
selecting an effective rectangular area of the image to be detected according to the sea antenna frame;
transversely dividing the effective rectangular region into N 2 The image block comprises image blocks, wherein an overlapping area exists between two adjacent image blocks, the transverse widths of all the image blocks are the same, and the value of N is a positive integer greater than one;
will N 2 Arranging the image blocks according to N rows and N columns to obtain a recombined image of the image to be detected;
detecting sea surface targets in the recombined image by using a pre-constructed deep learning network model, obtaining first coordinate information of each sea surface target, and combining the first coordinate information of each sea surface target into a first coordinate information set, wherein the first coordinate information is the coordinate information of the sea surface target in the recombined image;
and respectively converting the first coordinate information of each sea surface target into second coordinate information of the sea surface target, and combining the second coordinate information of each sea surface target into a second coordinate information set, wherein the second coordinate information is the coordinate information of the sea surface target in the image to be detected.
In a further improvement, after the step of respectively converting the first coordinate information of each sea surface target into the second coordinate information of the sea surface target and combining the second coordinate information of each sea surface target into the second coordinate information set, the method further includes the following steps:
and removing repeated sea surface target coordinate information in the second coordinate information set.
Further improved, the sea-sky line in the image to be detected is identified, specifically including:
calculating the vertical gradient of the image to be detected, and extracting to obtain edge features;
obtaining an edge straight line segment according to the edge characteristics;
screening the edge straight line segments according to a preset first threshold value to obtain target straight line segments;
aggregating the target straight line segments by adopting a preset clustering algorithm to obtain a sea-sky line segment set;
and fitting the sea-sky-line segment set by adopting a least square method to obtain the sea-sky-line.
Further improved, the selecting an effective rectangular area of the image to be detected according to the sea-sky-line frame specifically includes:
identifying the intersection point of the sea-sky-line and the left boundary line of the image to be detected, and recording the coordinates of the intersection point as
；
Identifying the intersection point of the sea-sky-line and the right boundary line of the image to be detected,and the coordinates of the intersection point are recorded as
；
Calculating the coordinates of the center point of the sea-sky-line according to the coordinates of the intersection point of the sea-sky-line and the left boundary line of the image to be detected and the coordinates of the intersection point of the sea-sky-line and the right boundary line of the image to be detected
Wherein
， 
；
Judgment equation
Whether the result is true or not; if yes, determining a first distance parameter
And a second distance parameter
If not, determining a first distance parameter
And a second distance parameter
Wherein W is the width of the image to be detected;
the coordinates of the upper boundary line are calculated from the first distance parameter d1
And calculating the coordinates of the lower boundary line based on the second distance parameter d2
；
And forming an effective rectangular area of the image to be detected by the upper boundary line, the lower boundary line, the left boundary line of the image to be detected and the right boundary line of the image to be detected.
In a further improvement, the first coordinate information of the sea surface object includes coordinate information of a center point of the sea surface object, height information of an object frame for framing the sea surface object, and width information of the object frame.
In a further improvement, the transverse widths of the overlapping regions are all first preset intervals, and the first preset intervals are positive integers.
In a further improvement, the calculating the vertical gradient of the image to be detected specifically includes:
calculating the vertical gradient of the image to be detected by a kernel operator, wherein
Weighted value
。
In a further improvement, the method further includes, after removing the repeated sea surface target coordinate information in the second coordinate information set, the following steps:
and removing the coordinate information of the interference target in the second coordinate information set to obtain a final coordinate information set of the sea surface target in the image to be detected, wherein the interference target comprises an airplane and a bird in the sky.
In a further improvement, the removing the coordinate information of the interference target in the second coordinate information set specifically includes:
calculating the vertical coordinate of the lower right corner of each sea surface target based on the second coordinate information set, and recording the vertical coordinate of the lower right corner as the vertical coordinate
；
Judgment of
If the sea surface target is not the sea surface target, keeping the coordinate information of the sea surface target in a second coordinate information set, and if the sea surface target is not the sea surface target, removing the coordinate information of the sea surface target from the second coordinate information set;
wherein the content of the first and second substances,
and m is the number of the remaining sea surface targets after the repeated sea surface target coordinate information in the second coordinate information set is removed.
The invention has the following beneficial effects:
1) the sea antenna in the image to be detected is fitted, an effective rectangular area of the image to be detected is selected based on the sea antenna, the effective rectangular area is transversely divided into a plurality of image blocks, meanwhile, in order to avoid cutting of a sea surface target, when the effective rectangular area is transversely divided, two adjacent image blocks are partially overlapped, all the image blocks are recombined, the recombined image is input into a pre-constructed deep learning network model to be subjected to target detection, and the detected coordinates of the sea surface target are converted into the coordinate position in the image to be detected.
According to the method, the interference of most invalid information is eliminated through the selection of the effective area, and meanwhile, compared with the original image to be detected, the resolution of the sea surface target input into the deep learning network model is increased by the recombined image, so that the accuracy of the detection of the sea surface target is improved, and the detection of small target ships, small target buoys and the like near the sea antenna under a complex sea antenna scene is completed.
2) And calculating the vertical gradient of the image to be detected through a kernel operator, and combining a least square method to realize the accurate simulation of the sea-sky-line.
3) And the precision of sea surface target detection is further improved by removing the repeated target and the interference target coordinate information in the second coordinate information set.
Drawings
FIG. 1 is a flow chart of a method for detecting small targets based on perspective transformation in a sea-sky-line scenario;
FIG. 2 is a schematic diagram of a division of an effective rectangular area;
fig. 3 is a schematic diagram of a recombined image obtained by recombining image blocks.
Detailed Description
The technical solutions of the present invention will be described clearly and completely with reference to the following embodiments, and it should be apparent that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without inventive effort based on the embodiments of the present invention, are within the scope of the present invention.
As shown in fig. 1, the present embodiment provides a method for detecting a small target based on view angle conversion in a sea-sky-line scene, which is used for detecting a small sea-surface target such as a small ship and a small buoy near a sea-sky-line, and specifically includes the following steps:
and S1, acquiring an image to be detected.
And S2, identifying the sea-sky-line in the image to be detected. In a common embodiment, before the sea-sky-line in the image to be detected is identified, filtering and denoising processing is further performed on the image to be detected.
In the present embodiment, S2 includes the following sub-steps:
and a substep S21 of calculating the vertical direction gradient of the image to be detected and extracting to obtain edge characteristics.
And a substep S22 of obtaining an edge straight line segment according to the edge characteristics.
And a substep S23 of screening the edge straight line segment according to a preset first threshold value to obtain a target straight line segment.
And a substep S24, adopting a preset k-meas clustering algorithm to aggregate the target straight line segments to obtain a sea-sky line segment set.
And a substep S25 of fitting the sea-sky-line segment set by adopting a least square method to obtain the sea-sky-line.
And S3, selecting an effective rectangular area of the image to be detected according to the sea antenna frame.
In the present embodiment, S3 includes the following sub-steps:
substep S31, identifying sea-sky-lineThe intersection point of the image and the left boundary line of the image to be detected, and the coordinate of the intersection point is recorded as
。
Substep S32, identifying the intersection point of the sea-sky-line and the right boundary line of the image to be detected, and recording the coordinates of the intersection point as
。
Substep S33, calculating coordinates of center point of sea-sky-line according to coordinates of intersection point of sea-sky-line and left boundary line of image to be detected and coordinates of intersection point of sea-sky-line and right boundary line of image to be detected
Wherein
， 
。
Substep S34, determining equation
Whether the result is true or not; if yes, determining a first distance parameter
And a second distance parameter
At this time, the first value mode of the first distance parameter d1 and the second distance parameter d2 is adopted; if not, determining a first distance parameter
And a second distance parameter
In this case, the first distance parameter d1 and the second distance parameter d2 have the second value. Wherein W is the width of the image to be detected. The step determines the value mode of the first distance parameter d1 and the second distance parameter d2 according to the calculation of the sea-sky-line inclination angle. When the sea-sky-line level or the inclination angle of the sea-sky-line is less than 10 degrees, the second value mode is adopted by the first distance parameter d1 and the second distance parameter d2, and when the inclination angle of the sea-sky-line is greater than or equal to 10 degrees, the first value mode is adopted by the first distance parameter d1 and the second distance parameter d 2. In the first value taking mode, the calculation coefficient 1.2 and the calculation coefficient 1.5 are preferable values based on experience, and can be adjusted according to specific situations, and meanwhile, the sea surface target close to the lens is relatively large, so that the distance parameter at the ocean side is set to be larger, the sea surface target far away from the lens side is relatively smaller, and the distance parameter at the sky side is set to be smaller. The calculation factor in the second distance parameter d2 is 1.5 and the calculation factor in the first distance parameter d1 is 1.2.
Substep S35, calculating coordinates of the upper boundary line based on the first distance parameter d1
And calculating the coordinates of the lower boundary line based on the second distance parameter d2
。
Substep S36, forming an effective rectangular region of the image to be detected by the upper boundary line, the lower boundary line, the left boundary line of the image to be detected and the right boundary line of the image to be detected.
In addition, the coordinates of the left boundary line of the image to be detected
Coordinates of the right boundary line of the image to be detected
. In this embodiment, it is preferable
Thus, therefore, it is
。
S4, the effective rectangular area is divided into N by N image blocks, an overlap area having a lateral width of a first preset interval d3 exists between any two adjacent image blocks, and the lateral widths of all the image blocks are the same, where the value of N is a positive integer greater than one, and the value of the first preset interval d3 is a positive integer.
The calculation of the lateral width w1 of an image block is as follows:
calculating a first intermediate parameter
N is a positive integer;
calculating the lateral width of an image block
W1 is a positive integer, specifically: take the positive integer closest to the actual calculated value of w1 and greater than the actual value of w 1.
The calculation process of N in the transverse division of the effective rectangular region is as follows:
if it is
If N is equal to 2; if it is
If N is 3; if it is
If so, then N takes the value of 4; if it is
Then N takes the value 5.
And S5, arranging the N-by-N image blocks according to N rows and N columns to obtain a recombined image of the image to be detected.
S6, inputting the recombined images into a pre-constructed deep learning network model for target detection, identifying sea surface targets in the recombined images, obtaining first coordinate information of the sea surface targets, and combining the obtained first coordinate information of the sea surface targets into a first coordinate information set. The first coordinate information is coordinate information of the sea surface target in the recombined image. The deep learning network model is used for detecting sea surface targets in a common embodiment. The first coordinate information of the sea surface target comprises the coordinate information of the center point of the sea surface target, the height information of a target frame for framing the sea surface target and the width information of the target frame.
And S7, converting the first coordinate information of each sea surface target into second coordinate information of the sea surface target respectively, and combining the second coordinate information of each sea surface target into a second coordinate information set. And the second coordinate information is the coordinate information of the sea surface target in the image to be detected. And the second coordinate information of the sea surface target comprises the coordinate information of the central point of the sea surface target in the image to be detected, the height information of a target frame for framing the sea surface target and the width information of the target frame.
Preferably, the following steps are also included after S7:
and S8, removing the repeated sea surface target coordinate information in the second coordinate information set. Because two adjacent image blocks have an overlapping area with the transverse width of the first preset interval d3, after the target detection is performed through the deep learning network model and then the coordinate conversion is performed, a plurality of sea surface targets with the same coordinate information may appear, and the precision of the sea surface target detection is improved by screening out the repeated sea surface target coordinate information.
Preferably, the vertical gradient of the image to be detected is calculated, specifically:
calculating the vertical gradient of the image to be detected by a kernel operator, wherein
Weighted value
. The larger the weight value k is, the larger the gradient value in the vertical direction is, and k takes a value of 1 in this embodiment.
Preferably, the following steps are also included after S8:
and S9, removing the coordinate information of the interference target in the second coordinate information set to obtain a final coordinate information set of the sea surface target in the image to be detected. The interference target includes an airplane, a bird and the like in the sky near the sea-sky.
Removing the coordinate information of the interference target in the second coordinate information set, which specifically comprises the following substeps:
s91, calculating the vertical coordinate of the lower right corner of each sea surface target based on the second coordinate information set, and recording the vertical coordinate of the lower right corner as
。
S92, judgment
And if not, removing the coordinate information of the sea surface target from the second coordinate information set.
Wherein the content of the first and second substances,
and m is the number of the remaining sea surface target coordinates after the repeated sea surface target coordinate information in the second coordinate information set is removed, namely the number of the remaining sea surface targets. And the accuracy of sea surface target detection is further improved by screening out the coordinate information of the interference target in the second coordinate information set.
The first preset interval d3 is preferably greater than half of the maximum width of the sea surface targets such as various small target vessels and small target buoys, based on the type of small target vessels and small target buoys to be detected.
With reference to fig. 2 and 3, the following describes a generation process of a reconstructed image (four image blocks are arranged in two rows and two columns in a rectangular manner) and a specific calculation process of converting each coordinate information in the first coordinate information set to a corresponding coordinate position in an image to be detected, by dividing the effective rectangular area into four parts by transverse division.
The generation process of the recombined image comprises the following steps:
four image blocks obtained by transversely dividing the effective rectangular area are sequentially defined as a first image block, a second image block, a third image block and a fourth image block from left to right;
placing a first image block in a first row and a first column of a rectangular arrangement, placing a second image block in a first row and a second column of the rectangular arrangement, placing a third image block in a second row and the first column of the rectangular arrangement, and placing a fourth image block in a second row and the second column of the rectangular arrangement to obtain a recombined image;
a specific calculation process for converting each piece of coordinate information in the first set of coordinate information to a corresponding coordinate position in the image to be detected:
converting the coordinate information of the sea surface target detected in the first image block in the recombined image into the coordinate position in the image to be detected, wherein the calculation formula is as follows:
；
converting the coordinate information of the sea surface target detected in the second image block in the recombined image into the coordinate position in the image to be detected, wherein the calculation formula is as follows:
converting the coordinate information of the sea surface target detected in the third image block in the recombined image into the coordinate position in the image to be detected, wherein the calculation formula is as follows:
converting the coordinate information of the sea surface target detected in the fourth image block in the recombined image into the coordinate position in the image to be detected, wherein the calculation formula is as follows:
；
wherein the content of the first and second substances,
the coordinates of the sea surface object in the recombined image,
for coordinates of sea surface objects detected by the reconstructed image in the image to be detected, and
。
the method for detecting the small target ship in the sea-sky-line scene adopts three means of selecting an effective rectangular area, transversely dividing an image to be detected and recombining image blocks, belongs to a conversion realization of a machine visual angle in deep learning, saves the operation amount of a deep learning network model, increases the resolution ratio of a sea surface small target in image data input into the deep learning network model, greatly improves the detection precision of the small target ship and a small target buoy near the sea-sky-line, overcomes the defects of the existing target detection scheme mentioned in the background technology, and has a larger application prospect.
The foregoing is illustrative of the preferred embodiments of this invention, and it is to be understood that the invention is not limited to the precise form disclosed herein and that various other combinations, modifications, and environments may be resorted to, falling within the scope of the concept as disclosed herein, either as described above or as apparent to those skilled in the relevant art. And that modifications and variations may be effected by those skilled in the art without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (9)
1. The method for detecting the small target based on the view angle conversion in the sea-sky-line scene is characterized by comprising the following steps of:
acquiring an image to be detected;
identifying a sea-sky line in an image to be detected;
selecting an effective rectangular area of the image to be detected according to the sea antenna frame;
transversely dividing the effective rectangular region into N 2 The image block comprises image blocks, wherein an overlapping area exists between two adjacent image blocks, the transverse widths of all the image blocks are the same, and the value of N is a positive integer greater than one;
will N 2 Arranging the image blocks according to N rows and N columns to obtain a recombined image of the image to be detected;
detecting sea surface targets in the recombined image by using a pre-constructed deep learning network model, obtaining first coordinate information of each sea surface target, and combining the first coordinate information of each sea surface target into a first coordinate information set, wherein the first coordinate information is the coordinate information of the sea surface target in the recombined image;
and respectively converting the first coordinate information of each sea surface target into second coordinate information of the sea surface target, and combining the second coordinate information of each sea surface target into a second coordinate information set, wherein the second coordinate information is the coordinate information of the sea surface target in the image to be detected.
2. The method for detecting small targets based on view angle transformation in sea-sky-line scene as claimed in claim 1, wherein the step of converting the first coordinate information of each sea-surface target into the second coordinate information of the sea-surface target and combining the second coordinate information of each sea-surface target into the second coordinate information set further comprises the following steps:
and removing repeated sea surface target coordinate information in the second coordinate information set.
3. The method for detecting the small target based on the view angle conversion in the sea-sky-line scene as claimed in claim 1, wherein the identifying the sea-sky-line in the image to be detected specifically comprises:
calculating the vertical gradient of the image to be detected, and extracting to obtain edge features;
obtaining an edge straight line segment according to the edge characteristics;
screening the edge straight line segments according to a preset first threshold value to obtain target straight line segments;
aggregating the target straight line segments by adopting a preset clustering algorithm to obtain a sea-sky line segment set;
and fitting the sea-sky-line segment set by adopting a least square method to obtain the sea-sky-line.
4. The method for detecting the small target based on the view angle conversion in the sea-sky-line scene as claimed in claim 2, wherein the selecting the effective rectangular area of the image to be detected according to the sea-sky-line frame specifically comprises:
identifying the intersection point of the sea-sky-line and the left boundary line of the image to be detected, and recording the coordinates of the intersection point as
；
Identifying the intersection point of the sea-sky-line and the right boundary line of the image to be detected, and recording the coordinates of the intersection point as
；
Calculating the coordinates of the center point of the sea-sky-line according to the coordinates of the intersection point of the sea-sky-line and the left boundary line of the image to be detected and the coordinates of the intersection point of the sea-sky-line and the right boundary line of the image to be detected
Wherein
，
；
Judging equality
Whether the result is true or not; if yes, determining a first distance parameter
And a second distance parameter
If not, determining a first distance parameter
And a second distance parameter
Wherein W is the width of the image to be detected;
the coordinates of the upper boundary line are calculated from the first distance parameter d1
And calculating the coordinates of the lower boundary line based on the second distance parameter d2
；
And forming an effective rectangular area of the image to be detected by the upper boundary line, the lower boundary line, the left boundary line of the image to be detected and the right boundary line of the image to be detected.
5. The method for detecting the small target based on the view angle conversion in the sea-sky-line scene according to claim 1, wherein the first coordinate information of the sea-surface target includes coordinate information of a center point of the sea-surface target, height information of a target frame for framing the sea-surface target, and width information of the target frame.
6. The method for detecting the small target based on the view angle conversion in the sea-sky-line scene as claimed in claim 1, wherein the horizontal widths of the overlapping regions are all a first preset interval, and the first preset interval is a positive integer.
7. The method for detecting the small target based on the view angle conversion in the sea-sky-line scene as claimed in claim 3, wherein the calculating the vertical gradient of the image to be detected specifically comprises:
calculating the vertical gradient of the image to be detected by a kernel operator, wherein
Weighted value
。
8. The method for detecting small targets based on view angle conversion in sea-sky-line scene as claimed in claim 4, wherein the step of removing the repeated sea-surface target coordinate information in the second coordinate information set further comprises the following steps:
and removing the coordinate information of the interference target in the second coordinate information set to obtain a final coordinate information set of the sea surface target in the image to be detected, wherein the interference target comprises an airplane and a bird in the sky.
9. The method for detecting the small target based on the perspective transformation in the sea-sky-line scene according to claim 8, wherein the removing the coordinate information of the interfering target in the second coordinate information set specifically includes:
calculating the vertical coordinate of the lower right corner of each sea surface target based on the second coordinate information set, and recording the vertical coordinate of the lower right corner as the vertical coordinate
；
Judgment of
If the sea surface target is not the sea surface target, keeping the coordinate information of the sea surface target in a second coordinate information set, and if the sea surface target is not the sea surface target, removing the coordinate information of the sea surface target from the second coordinate information set;
wherein the content of the first and second substances,
and m is the number of the remaining sea surface targets after the repeated sea surface target coordinate information in the second coordinate information set is removed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210786036.1A CN114863258B (en) | 2022-07-06 | 2022-07-06 | Method for detecting small target based on visual angle conversion in sea-sky-line scene |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210786036.1A CN114863258B (en) | 2022-07-06 | 2022-07-06 | Method for detecting small target based on visual angle conversion in sea-sky-line scene |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114863258A true CN114863258A (en) | 2022-08-05 |
| CN114863258B CN114863258B (en) | 2022-09-06 |
Family
ID=82625993
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210786036.1A Active CN114863258B (en) | 2022-07-06 | 2022-07-06 | Method for detecting small target based on visual angle conversion in sea-sky-line scene |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114863258B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115049907A (en) * | 2022-08-17 | 2022-09-13 | 四川迪晟新达类脑智能技术有限公司 | FPGA-based YOLOV4 target detection network implementation method |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103679694A (en) * | 2013-05-29 | 2014-03-26 | 哈尔滨工程大学 | Ship small target detection method based on panoramic vision |
| CN104599273A (en) * | 2015-01-22 | 2015-05-06 | 南京理工大学 | Wavelet multi-scale crossover operation based sea-sky background infrared small target detection method |
| CN108229342A (en) * | 2017-12-18 | 2018-06-29 | 西南技术物理研究所 | A kind of surface vessel target automatic testing method |
| CN108846844A (en) * | 2018-04-13 | 2018-11-20 | 上海大学 | A kind of sea-surface target detection method based on sea horizon |
| CN110188696A (en) * | 2019-05-31 | 2019-08-30 | 华南理工大学 | A kind of water surface is unmanned to equip multi-source cognitive method and system |
| CN111091024A (en) * | 2018-10-23 | 2020-05-01 | 广州弘度信息科技有限公司 | Small target filtering method and system based on video recognition result |
| CN111767856A (en) * | 2020-06-29 | 2020-10-13 | 哈工程先进技术研究院(招远)有限公司 | Infrared small target detection algorithm based on gray value statistical distribution model |
| CN112258518A (en) * | 2020-10-09 | 2021-01-22 | 国家海洋局南海调查技术中心(国家海洋局南海浮标中心) | Sea-sky-line extraction method and device |
| CN112669332A (en) * | 2020-12-28 | 2021-04-16 | 大连海事大学 | Method for judging sea and sky conditions and detecting infrared target based on bidirectional local maximum and peak local singularity |
| CN113223000A (en) * | 2021-04-14 | 2021-08-06 | 江苏省基础地理信息中心 | Comprehensive method for improving small target segmentation precision |
| US20210256680A1 (en) * | 2020-02-14 | 2021-08-19 | Huawei Technologies Co., Ltd. | Target Detection Method, Training Method, Electronic Device, and Computer-Readable Medium |
| CN114494179A (en) * | 2022-01-24 | 2022-05-13 | 深圳闪回科技有限公司 | Mobile phone back damage point detection method and system based on image recognition |
-
2022
- 2022-07-06 CN CN202210786036.1A patent/CN114863258B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103679694A (en) * | 2013-05-29 | 2014-03-26 | 哈尔滨工程大学 | Ship small target detection method based on panoramic vision |
| CN104599273A (en) * | 2015-01-22 | 2015-05-06 | 南京理工大学 | Wavelet multi-scale crossover operation based sea-sky background infrared small target detection method |
| CN108229342A (en) * | 2017-12-18 | 2018-06-29 | 西南技术物理研究所 | A kind of surface vessel target automatic testing method |
| CN108846844A (en) * | 2018-04-13 | 2018-11-20 | 上海大学 | A kind of sea-surface target detection method based on sea horizon |
| CN111091024A (en) * | 2018-10-23 | 2020-05-01 | 广州弘度信息科技有限公司 | Small target filtering method and system based on video recognition result |
| CN110188696A (en) * | 2019-05-31 | 2019-08-30 | 华南理工大学 | A kind of water surface is unmanned to equip multi-source cognitive method and system |
| US20210256680A1 (en) * | 2020-02-14 | 2021-08-19 | Huawei Technologies Co., Ltd. | Target Detection Method, Training Method, Electronic Device, and Computer-Readable Medium |
| CN111767856A (en) * | 2020-06-29 | 2020-10-13 | 哈工程先进技术研究院(招远)有限公司 | Infrared small target detection algorithm based on gray value statistical distribution model |
| CN112258518A (en) * | 2020-10-09 | 2021-01-22 | 国家海洋局南海调查技术中心(国家海洋局南海浮标中心) | Sea-sky-line extraction method and device |
| CN112669332A (en) * | 2020-12-28 | 2021-04-16 | 大连海事大学 | Method for judging sea and sky conditions and detecting infrared target based on bidirectional local maximum and peak local singularity |
| CN113223000A (en) * | 2021-04-14 | 2021-08-06 | 江苏省基础地理信息中心 | Comprehensive method for improving small target segmentation precision |
| CN114494179A (en) * | 2022-01-24 | 2022-05-13 | 深圳闪回科技有限公司 | Mobile phone back damage point detection method and system based on image recognition |
Non-Patent Citations (4)
| Title |
|---|
| YUNZHONG HOU等: "《Multiview Detection with Feature Perspective Transformation》", 《ECCV》 * |
| 孙维亚 等: "《融合运动特征的高效视频火焰检测算法》", 《数据采集与处理》 * |
| 徐海祥 等: "《水面图像目标检测的强语义特征提取结构》", 《华中科技大学学报(自然科学版)》 * |
| 王啸雨: "《基于深度学习的水面无人艇船舶检测技术研究》", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115049907A (en) * | 2022-08-17 | 2022-09-13 | 四川迪晟新达类脑智能技术有限公司 | FPGA-based YOLOV4 target detection network implementation method |
| CN115049907B (en) * | 2022-08-17 | 2022-10-28 | 四川迪晟新达类脑智能技术有限公司 | FPGA-based YOLOV4 target detection network implementation method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114863258B (en) | 2022-09-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108961235B (en) | Defective insulator identification method based on YOLOv3 network and particle filter algorithm | |
| CN109934826B (en) | Image feature segmentation method based on graph convolution network | |
| CN108648161B (en) | Binocular vision obstacle detection system and method of asymmetric kernel convolution neural network | |
| CN108960135B (en) | Dense ship target accurate detection method based on high-resolution remote sensing image | |
| CN107256557A (en) | A kind of controllable subdivision curved surface image vector method of error | |
| CN110084241B (en) | Automatic ammeter reading method based on image recognition | |
| CN113435282B (en) | Unmanned aerial vehicle image ear recognition method based on deep learning | |
| CN111723464A (en) | Typhoon elliptic wind field parametric simulation method based on remote sensing image characteristics | |
| CN109087261A (en) | Face antidote based on untethered acquisition scene | |
| CN114863258B (en) | Method for detecting small target based on visual angle conversion in sea-sky-line scene | |
| CN108460833A (en) | A kind of information platform building traditional architecture digital protection and reparation based on BIM | |
| CN103839234A (en) | Double-geometry nonlocal average image denoising method based on controlled nuclear | |
| CN109389553B (en) | Meteorological facsimile picture contour interpolation method based on T spline | |
| CN115861409B (en) | Soybean leaf area measuring and calculating method, system, computer equipment and storage medium | |
| CN112906689A (en) | Image detection method based on defect detection and segmentation depth convolution neural network | |
| CN107944497A (en) | Image block method for measuring similarity based on principal component analysis | |
| CN109241981B (en) | Feature detection method based on sparse coding | |
| CN114742864A (en) | Belt deviation detection method and device | |
| CN102324043B (en) | Image matching method based on DCT (Discrete Cosine Transformation) through feature description operator and optimization space quantization | |
| CN114627461A (en) | Method and system for high-precision identification of water gauge data based on artificial intelligence | |
| CN113946978A (en) | Underwater three-dimensional temperature and salinity parallel forecasting method based on LightGBM model | |
| CN116721228B (en) | Building elevation extraction method and system based on low-density point cloud | |
| CN113344148A (en) | Marine ship target identification method based on deep learning | |
| CN113658144A (en) | Method, device, equipment and medium for determining pavement disease geometric information | |
| CN108734148A (en) | A kind of public arena image information collecting unmanned aerial vehicle control system based on cloud computing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |