CN113537089A - Pine wood nematode disease infected wood identification and positioning method based on unmanned aerial vehicle aerial photography original sheet - Google Patents
Pine wood nematode disease infected wood identification and positioning method based on unmanned aerial vehicle aerial photography original sheet Download PDFInfo
- Publication number
- CN113537089A CN113537089A CN202110820396.4A CN202110820396A CN113537089A CN 113537089 A CN113537089 A CN 113537089A CN 202110820396 A CN202110820396 A CN 202110820396A CN 113537089 A CN113537089 A CN 113537089A
- Authority
- CN
- China
- Prior art keywords
- wood
- epidemic
- image
- original
- model
- 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.)
- Pending
Links
- 239000002023 wood Substances 0.000 title claims abstract description 83
- 238000000034 method Methods 0.000 title claims abstract description 37
- 241000243771 Bursaphelenchus xylophilus Species 0.000 title claims abstract description 26
- 201000010099 disease Diseases 0.000 title claims abstract description 26
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 title claims abstract description 26
- 241000196324 Embryophyta Species 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims description 54
- 238000001514 detection method Methods 0.000 claims description 18
- 235000008331 Pinus X rigitaeda Nutrition 0.000 claims description 9
- 235000011613 Pinus brutia Nutrition 0.000 claims description 9
- 241000018646 Pinus brutia Species 0.000 claims description 9
- 238000013135 deep learning Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 9
- 238000005516 engineering process Methods 0.000 claims description 8
- 238000002372 labelling Methods 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 4
- 241000607479 Yersinia pestis Species 0.000 claims description 3
- 239000007787 solid Substances 0.000 claims description 3
- 238000009432 framing Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 241000894007 species Species 0.000 description 3
- 230000003321 amplification Effects 0.000 description 2
- 238000013434 data augmentation Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000003199 nucleic acid amplification method Methods 0.000 description 2
- 241000244206 Nematoda Species 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000011347 resin Substances 0.000 description 1
- 229920005989 resin Polymers 0.000 description 1
- 230000028327 secretion Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/5866—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, manually generated location and time information
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/587—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using geographical or spatial information, e.g. location
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Library & Information Science (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of pine wood nematode disease image identification and positioning, and discloses a pine wood nematode disease epidemic wood identification and positioning method based on an unmanned aerial vehicle aerial photograph original sheet, wherein after the epidemic wood is identified on the original sheet, a self-designed special mark is embedded in the position of the epidemic wood, the special mark has obvious appearance characteristics in a mountain forest and is not easy to be confused with other objects, and the characteristics can be still kept after orthographic images are spliced; the special marks are detected and identified on the orthographic images with accurate longitude and latitude information, and the longitude and latitude position of each special mark can be accurately obtained, so that the quantity and distribution of the wood epidemic in the target mountain forest area range are accurately counted, and the identification and positioning of the wood epidemic caused by the pine wood nematode disease of the plant are accurately achieved.
Description
Technical Field
The invention relates to the technical field of pine wood nematode disease image identification and positioning, in particular to a pine wood nematode disease wood identification and positioning method based on an unmanned aerial vehicle aerial photography original film.
Background
Pine wilt disease, also known as pine wilt disease, is a devastating forest disease caused by pine, and pine infected with the pine has yellow brown or red brown needle, wilting and drooping, stopped resin secretion, gradual withering and death, and finally rotting.
Because the diseased pine wood nematode epidemic trees are yellow brown or red brown, coniferous leaves gradually wither, and the appearance of the disease nematode epidemic trees is different from that of normal pine trees and other tree species in mountain forests, the current scheme for acquiring mountain forest images by using an unmanned aerial vehicle aerial photography technology and intelligently identifying the epidemic trees based on a deep learning technology also exists.
There are mainly two types of existing solutions:
(1) the epidemic wood recognition is directly carried out on the original aerial photograph of the unmanned aerial vehicle. According to the scheme, the unmanned aerial vehicle is used for obtaining aerial images of the target mountain forest area, and various deep learning algorithm models are used for directly detecting and identifying the pine wood nematode disease on aerial image original sheets.
And directly carry out the shortcoming of epidemic wood recognition technical scheme on the unmanned aerial vehicle original film of taking photo by plane: unmanned aerial vehicle is when gathering mountain forest image, the longitude and latitude information of the position that unmanned aerial vehicle was located when this original film was shot also can be acquireed according to GPS signal to the sola original film of shooting, however owing to receive unmanned aerial vehicle flight angle course planning and the influence of wind direction at that time, the positive central point position of this image can't pinpoint to this longitude and latitude coordinate, simultaneously because mountain forest topography fluctuation is uneven, can't be accurate carry out the one-to-one with every pixel position and longitude and latitude coordinate in the original film, be difficult to carry out accurate longitude and latitude location to the epidemic wood of discerning, can't accurate matching deduplication in the adjacent original film with the epidemic wood of the same strain, consequently although the epidemic wood discernment effect is better, but can't accomplish to carry out accurate quantity and distribution statistics to the epidemic wood in the regional scope.
(2) And carrying out epidemic wood identification and positioning on the orthographic picture which is attached with longitude and latitude coordinate information after splicing. According to the scheme, firstly, an unmanned aerial vehicle is used for collecting aerial images of a mountain forest, large overlapping needs to be achieved between adjacent aerial image original sheets, phase control points are arranged on the ground at intervals, after the aerial image original sheets of the unmanned aerial vehicle are obtained, the original sheets are spliced into large-range mountain forest area orthographic images with longitude and latitude coordinate information by means of image splicing software or algorithm, each pixel point in the orthographic images can obtain accurate longitude and latitude coordinate information, then, pine wood nematode disease epidemic trees are identified and positioned on the spliced orthographic images by means of a deep learning algorithm model, and the quantity and the distribution condition of the pine wood nematode disease epidemic trees in the target mountain forest range are counted.
The technical scheme for identifying and positioning the spliced orthographic image sheets has the following defects that: at the in-process that matches the concatenation with the original film of unmanned aerial vehicle aerial photography, because the shooting position of adjacent original film is different, there is visual angle distortion, the matching is accurate inadequately, the leaf that outside wind direction leads to shakes the influence such as, the orthoimage that the concatenation produced exists ghost, phenomenon such as fuzzy, the original film that the unmanned aerial vehicle shot of image quality relatively has the decline of certain degree, lead to this type of scheme to carry out accurate longitude and latitude location to each pixel position in the image, but there is great influence to the discernment accuracy of epidemic trees.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides a pine wood nematode infected wood identification and positioning method based on an unmanned aerial vehicle aerial photographic original sheet.
The technical scheme adopted by the invention for solving the technical problems is as follows: a pine wood nematode disease wood identification and positioning method based on an unmanned aerial vehicle aerial photography original film comprises the following steps:
(1) acquiring aerial image original sheets in a certain mountain forest area range in an aerial shooting mode by carrying a camera by an unmanned aerial vehicle, selecting an image partially containing pine wire pest epidemic trees as an epidemic tree training sample, and labeling the epidemic trees in the image of the epidemic tree training sample by an image labeling tool;
(2) selecting a proper object detection model based on a deep learning technology as an epidemic wood recognition model, and training the epidemic wood recognition model by using the epidemic wood training sample generated in the step (1);
(3) performing epidemic wood detection and identification on the images of all aerial image original films by using the epidemic wood identification model trained in the step (2), and acquiring all epidemic wood and position information thereof in the image of each aerial image original film;
(4) embedding special marks with obvious characteristics at the epidemic wood positions of the images of the aerial image original films obtained in the step (3), and splicing all the aerial image original films embedded with the special marks into an ortho-image with longitude and latitude coordinates by using image splicing software or an algorithm;
(5) selecting an image containing a special mark in the orthographic image as a special mark training sample, and marking the special mark in the special mark training sample image by an image marking tool;
(6) selecting a proper object detection model based on the deep learning technology as a special mark recognition model, and training the special mark recognition model by using the special mark training sample generated in the step (5);
(7) performing special mark recognition on all the orthoimages by using the special mark recognition model trained in the step (6) to obtain all special marks and position information thereof in the orthoimages;
(8) and (4) converting the coordinate information of the special mark generated in the step (7) into longitude and latitude coordinates, acquiring the quantity of the epidemic trees in the range of the target mountain forest and the longitude and latitude of each epidemic tree, and realizing the identification and positioning of the epidemic trees accurate to the plants.
Preferably, the method for obtaining the epidemic wood training samples in the step (1) comprises the steps of determining a range of a target mountain forest area in advance, planning and setting a flight path, setting a plurality of phase control points on the ground, enabling the unmanned aerial vehicle to fly at a constant speed at a fixed altitude, shooting one mountain forest area original sheet by the unmanned aerial vehicle carrying a camera at fixed time intervals to obtain all original sheets, uniformly cutting each original sheet into small images with 768 pixels, randomly selecting the small images cut out from part of the original sheets, and manually framing all the epidemic wood in the images by using an image labeling tool labelImg to obtain the epidemic wood training samples.
Preferably, in the log identification model in the step (2), the YOLOv4 object detection model is selected as a reference model for log identification, the value of the last full-connection layer of the YOLOv4 object detection model is set to be 2, the value is used for representing two types of new logs in the current year and previous logs in the previous year, and all log training samples are calculated according to the following formula (4): 1: 1, calculating a loss function through the training set, performing back propagation to update model parameters, tuning each hyper-parameter of the model through the verification set, and finally testing the identification accuracy of the model under the training sample data set of the log by using the test set.
Preferably, the method for detecting and identifying the epidemic wood in the step (3) includes the steps of splitting all aerial image original sheets shot by the unmanned aerial vehicle, uniformly splitting each original sheet into 768 × 768 pixels of small images, inputting the small images into a trained epidemic wood identification model for identifying the epidemic wood, outputting the pixel positions, the species and the confidence coefficient of the epidemic wood identification frame in the small images by the epidemic wood identification model, and finally converting the positions of all the epidemic wood identification frames belonging to the original sheet into the positions of the epidemic wood identification frames in the original sheet according to the positions of the small images in the original sheet.
Preferably, the distinctive special mark of the embedding feature in the step (4) is a white solid circle with a radius of 20 pixels.
Preferably, in the step (4), all original sheets embedded with the special marks are guided into image splicing software or algorithm for splicing to generate an orthoimage with longitude and latitude coordinates, and each pixel point in the orthoimage corresponds to one longitude and latitude coordinate.
Preferably, the special mark training sample in step (5) is obtained by dividing the obtained ortho-image into small graphs of 768 × 768 pixels, and randomly selecting a plurality of small graphs containing special marks as the special mark training samples, wherein the image marking tool is a labelImg marking tool.
Preferably, the special mark recognition model in step (6) selects the YOLOv4 model as the special mark recognition model, and the value of the last full connection layer of the special mark recognition model is set to 1.
Preferably, the method for performing special mark recognition on all the ortho images in the step (7) includes inputting the segmented ortho image small images into a trained special mark recognition model for recognition, outputting the pixel positions and confidence degrees of the special marks in the small images, namely the pixel positions of the epidemic trees in each small image, with the confidence degree threshold set to be 0.8, and finally keeping the recognition results of the special marks with the confidence degrees exceeding the threshold.
Preferably, the method for converting into longitude and latitude coordinates in step (8) includes obtaining pixel positions of center points of all the special marks in the ortho image according to the pixel positions of the minigrams in the ortho image and the pixel positions of the special marks in each minigram, and obtaining the longitude and latitude coordinates of the center points according to the corresponding relationship between the pixel positions and the longitude and latitude coordinates.
Compared with the prior art, the invention has the beneficial effects that:
1. the wood epidemic detection and identification are directly carried out on the original piece of the aerial image of the unmanned aerial vehicle, and the identification accuracy rate of the wood epidemic is higher due to the higher image quality of the original piece.
2. After the epidemic trees are identified on the original piece, the special mark which is designed by self is embedded in the position of the epidemic trees, the special mark has obvious appearance characteristics in mountain forests and is not easy to be confused with other objects, and the characteristics can be still kept after the orthographic images are spliced.
3. The special marks are detected and identified on the orthographic images with accurate longitude and latitude information, and the longitude and latitude position of each special mark can be accurately obtained, so that the quantity and distribution of the wood epidemic in the target mountain forest area range are accurately counted, and the identification and positioning of the wood epidemic caused by the pine wood nematode disease of the plant are accurately achieved.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2 is a schematic diagram of wood epidemic detection and identification of an aerial image original image.
Fig. 3 is a schematic diagram of an orthographic image spliced after embedding a special mark.
Fig. 4 is a schematic diagram of the obtained longitude and latitude coordinate data information.
Detailed Description
The following further describes embodiments of the present invention with reference to the drawings. It should be noted that the description of the embodiments is provided to help understanding of the present invention, but the present invention is not limited thereto. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Please refer to fig. 1, which illustrates a pine wood nematode disease wood identification and positioning method based on an original aerial photograph of an unmanned aerial vehicle, comprising the following steps:
(1) acquiring aerial image original sheets in a certain mountain forest area range in an aerial shooting mode by carrying a camera by an unmanned aerial vehicle, selecting an image partially containing pine wire pest epidemic trees as an epidemic tree training sample, and labeling the epidemic trees in the image of the epidemic tree training sample by an image labeling tool;
(2) selecting a proper object detection model based on a deep learning technology as an epidemic wood recognition model, and training the epidemic wood recognition model by using the epidemic wood training sample generated in the step (1);
(3) performing epidemic wood detection and identification on the images of all aerial image original films by using the epidemic wood identification model trained in the step (2), and acquiring all epidemic wood and position information thereof in the image of each aerial image original film;
(4) embedding special marks with obvious characteristics at the epidemic wood positions of the images of the aerial image original films obtained in the step (3), and splicing all the aerial image original films embedded with the special marks into an ortho-image with longitude and latitude coordinates by using image splicing software or an algorithm;
(5) selecting an image containing a special mark in the orthographic image as a special mark training sample, and marking the special mark in the special mark training sample image by an image marking tool;
(6) selecting a proper object detection model based on the deep learning technology as a special mark recognition model, and training the special mark recognition model by using the special mark training sample generated in the step (5);
(7) performing special mark recognition on all the orthoimages by using the special mark recognition model trained in the step (6) to obtain all special marks and position information thereof in the orthoimages;
(8) and (4) converting the coordinate information of the special mark generated in the step (7) into longitude and latitude coordinates, acquiring the quantity of the epidemic trees in the range of the target mountain forest and the longitude and latitude of each epidemic tree, and realizing the identification and positioning of the epidemic trees accurate to the plants.
The method for acquiring the epidemic wood training sample in the step (1) comprises the steps of predetermining a target mountain forest area range, planning and setting a flight path, setting a plurality of phase control points on the ground, enabling an unmanned aerial vehicle to fly at a constant speed at a fixed altitude, enabling the unmanned aerial vehicle to carry a camera to shoot a mountain forest area original sheet at fixed intervals to acquire all original sheets, and enabling subsequent orthographic images to be spliced to have high precision, enabling adjacent original sheets to have a large course overlapping rate and enabling adjacent flight paths to have a certain lateral overlapping rate; after all the original sheets are obtained, uniformly cutting each original sheet into small graphs with 768 pixels by 768 pixels, then randomly selecting the small graphs cut from part of the original sheets, and using an image marking tool.
The wood epidemic identification model in the step (2) selects a YOLOv4 object detection model with high identification accuracy and high identification speed as a reference model for wood epidemic identification, and in order to enable wood epidemic training samples to have diversity and wider representativeness, a training picture is subjected to a data augmentation strategy based on color space change and shape change, and the data augmentation strategy mainly comprises random change in contrast and saturation, random rotation, random left-right turning, random up-down turning and random scale scaling; because the epidemic wood training samples are classified into two types of the current new epidemic wood and the previous epidemic wood, the value of the last full connecting layer of the Yolov4 object detection model is set to be 2, the full connecting layer is used for representing two types of the current new epidemic wood and the previous epidemic wood, and all the epidemic wood training samples are classified into 4: 1: 1, calculating a loss function through the training set, performing back propagation to update model parameters, tuning each hyper-parameter of the model through the verification set, and finally testing the identification accuracy of the log identification model under the log training sample data set by using the test set.
As shown in fig. 2, the method for detecting and identifying epidemic wood in step (3) includes splitting all aerial image original sheets shot by the unmanned aerial vehicle, uniformly splitting each original sheet into 768 × 768 pixels of small images, inputting the small images into a trained epidemic wood identification model for identifying the epidemic wood, outputting the pixel positions, the species and the confidence degrees of the epidemic wood identification frame in the small images by the epidemic wood identification model, and finally converting the positions of all the epidemic wood identification frames belonging to the original sheet into the positions of the epidemic wood identification frame in the original sheet according to the positions of each small image in the original sheet.
Referring to the schematic diagram of the orthographic images spliced after the special marks are embedded in the orthographic images shown in fig. 3, in the step (4), the special marks with obvious characteristics are embedded in the positions of the epidemic trees of the images of the original sheets of each aerial image, because the overlapped parts of the adjacent original sheets are more, a strain of epidemic trees is repeatedly identified for many times, and because a single original sheet cannot be accurately positioned in the longitude and latitude, the duplication removal on the original sheet is difficult to directly carry out; after identifying the pixel position of the log on the original sheet, embedding a special mark into the central points of all identification frames; the selection of the special marker takes into account: firstly, the characteristics are obvious in mountain areas, and the appearance of the mountain areas is obviously different from the appearance of various objects; secondly, even if various stretching deformation and truncation occur to the special mark caused in the process of splicing the original sheets into the orthographic image, the special mark can still be normally identified; thirdly, the embedded special mark itself can not have obvious influence on image registration and splicing. Considering the above 3 points, the special mark with obvious embedding characteristics is selected as a white solid circle with a radius of 20 pixels, and it should be noted that any other object meeting the above three requirements can be selected as the special mark. And guiding all original sheets embedded with the special marks into image splicing software or an algorithm for splicing to generate an orthoimage with longitude and latitude coordinates, wherein the range covered by the orthoimage can reach hundreds of square kilometers, the pixel value can reach hundreds of billions of pixels, and each pixel point in the orthoimage corresponds to one longitude and latitude coordinate.
The special mark training sample in the step (5) is obtained by dividing the obtained ortho image into small images with 768 × 768 pixels, and randomly selecting a plurality of small images containing special marks as the special mark training samples, because distortion correction to the original sheet possibly exists in the process of splicing the ortho image, a certain epidemic strain is cut into different small images in the process of cutting the original sheet small images, the special marks at the splicing position have certain degree of deformation and incompleteness, and various scenes are contained as much as possible when the marked sample is selected; the image marking tool is a labelImg marking tool, and all special marks in the sample picture are manually selected.
Selecting a YOLOv4 model as the special mark recognition model in the step (6), wherein in the step (5), the special mark has shape change in the splicing process, data amplification of the image is carried out before training, and the amplification strategy mainly comprises random rotation, random left-right turning, random up-down turning and random stretching; because the special mark only selects one type, the value of the last full-connection layer of the special mark identification model is set as 1; the data set division mode and the model parameter training process are similar to the training log model in the step (2), namely, the data set division mode and the model parameter training process are divided into a training set, a verification set and a test set, the loss function is calculated through the training set, the model parameters are updated through back propagation, each hyper-parameter of the model is adjusted and optimized through the verification set, and finally the test set is used for testing the identification accuracy of the special mark identification model under the special mark training sample data set.
The method for recognizing the special marks of all the ortho images in the step (7) comprises the steps of inputting the segmented ortho image small images into a trained special mark recognition model for recognition, outputting the pixel positions and confidence degrees of the special marks in the small images, namely the pixel positions of the epidemic wood in each small image by the special mark recognition model, setting the confidence degree threshold value to be 0.8, finally keeping the recognition results of the special marks with the confidence degrees exceeding the threshold value, and reducing the false recognition rate of the special mark recognition model.
The method for converting the longitude and latitude coordinates in the step (8) comprises the steps of obtaining pixel positions of all special mark central points in the orthographic images according to the pixel positions of the small images in the orthographic images and the pixel positions of the special marks in each small image, obtaining the longitude and latitude coordinates of the central points through the corresponding relation between the pixel positions and the longitude and latitude coordinates, obtaining the number of the epidemic trees in a target mountain forest range and the longitude and latitude coordinate distribution of each plant of the epidemic trees, and realizing the identification and the positioning of the accurate plant of the plant, wherein as shown in fig. 4, each row of data in the diagram comprises 4 items of information, each item of information is separated by a space symbol and is respectively an epidemic tree type from left to right (1 represents a new epidemic tree in the current year, and 2 represents the longitude, latitude and the confidence coefficient, wherein the higher the confidence coefficient represents that the possibility that the model considers the plant is the epidemic trees is higher.
According to the method, the log detection and identification are directly carried out on the original image of the aerial image of the unmanned aerial vehicle, and the image quality of the original image is high, so that the log identification accuracy is high; after the epidemic trees are identified on the original piece, a self-designed special mark is embedded in the position of the epidemic trees, the special mark has obvious appearance characteristics in mountain forests and is not easy to be confused with other objects, and the characteristics can be still kept after the orthographic images are spliced; the special marks are detected and identified on the orthographic images with accurate longitude and latitude information, and the longitude and latitude position of each special mark can be accurately obtained, so that the quantity and distribution of the wood epidemic in the target mountain forest area range are accurately counted, and the identification and positioning of the wood epidemic caused by the pine wood nematode disease of the plant are accurately achieved.
The embodiments of the present invention have been described in detail with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. It will be apparent to those skilled in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, and the scope of protection is still within the scope of the invention.
Claims (10)
1. The pine wood nematode disease wood identification and positioning method based on the original aerial photography of the unmanned aerial vehicle is characterized by comprising the following steps of:
(1) acquiring aerial image original sheets in a certain mountain forest area range in an aerial shooting mode by carrying a camera by an unmanned aerial vehicle, selecting an image partially containing pine wire pest epidemic trees as an epidemic tree training sample, and labeling the epidemic trees in the image of the epidemic tree training sample by an image labeling tool;
(2) selecting a proper object detection model based on a deep learning technology as an epidemic wood recognition model, and training the epidemic wood recognition model by using the epidemic wood training sample generated in the step (1);
(3) performing epidemic wood detection and identification on the images of all aerial image original films by using the epidemic wood identification model trained in the step (2), and acquiring all epidemic wood and position information thereof in the image of each aerial image original film;
(4) embedding special marks with obvious characteristics at the epidemic wood positions of the images of the aerial image original films obtained in the step (3), and splicing all the aerial image original films embedded with the special marks into an ortho-image with longitude and latitude coordinates by using image splicing software or an algorithm;
(5) selecting an image containing a special mark in the orthographic image as a special mark training sample, and marking the special mark in the special mark training sample image by an image marking tool;
(6) selecting a proper object detection model based on the deep learning technology as a special mark recognition model, and training the special mark recognition model by using the special mark training sample generated in the step (5);
(7) performing special mark recognition on all the orthoimages by using the special mark recognition model trained in the step (6) to obtain all special marks and position information thereof in the orthoimages;
(8) and (4) converting the coordinate information of the special mark generated in the step (7) into longitude and latitude coordinates, acquiring the quantity of the epidemic trees in the range of the target mountain forest and the longitude and latitude of each epidemic tree, and realizing the identification and positioning of the epidemic trees accurate to the plants.
2. The method for identifying and positioning the pine wood nematode infected trees based on the original film aerial photography of the unmanned aerial vehicle according to claim 1, wherein the method for obtaining the tree training samples in the step (1) comprises the steps of predetermining the range of a target mountain forest region, planning and setting a route, setting a plurality of phase control points on the ground, enabling the unmanned aerial vehicle to fly at a constant speed at a fixed altitude, enabling the unmanned aerial vehicle to carry a camera to shoot one mountain forest region original film at fixed time intervals to obtain all original films, uniformly cutting each original film into small images with 768-768 pixels, randomly selecting small images cut out from part of the original films, and manually framing all the tree in the images by using an image labeling tool labelImg to obtain the tree training samples.
3. The method for identifying and positioning the pine wood nematode disease trees based on the original aerial photograph of the unmanned aerial vehicle as claimed in claim 1, wherein the tree identification model in the step (2) selects a YOLOv4 object detection model as a reference model for tree identification, the value of the last full connection layer of the YOLOv4 object detection model is set to be 2, the value is used for representing two tree types of new trees and former trees in the current year, and all tree training samples are set to be 4: 1: 1, calculating a loss function through the training set, performing back propagation to update model parameters, tuning each hyper-parameter of the model through the verification set, and finally testing the identification accuracy of the model under the training sample data set of the log by using the test set.
4. The pine wood nematode infected wood recognizing and positioning method based on the unmanned aerial vehicle aerial photography original sheet according to claim 1, characterized in that the method for detecting and recognizing the infected wood in the step (3) comprises the steps of segmenting all aerial photography image original sheets shot by the unmanned aerial vehicle, uniformly segmenting each original sheet into 768 × 768 pixel small graphs, inputting the small graphs into a trained infected wood recognizing model for infected wood recognition, outputting the pixel positions of the infected wood recognizing frames in the small graphs, the kinds and the confidence degrees of the infected wood by the infected wood recognizing model, and finally converting the positions of all the infected wood recognizing frames belonging to the original sheet into the positions of the original sheets according to the positions of each small graph in the original sheet.
5. The method for identifying and positioning the pine wood nematode disease trees based on the original aerial photograph of the unmanned aerial vehicle as claimed in claim 1, wherein the special mark with obvious embedding characteristics in the step (4) is a white solid circle with a radius of 20 pixels.
6. The method for identifying and positioning the pine wood nematode disease trees based on the original film aerial-photography by the unmanned aerial vehicle according to claim 1, wherein in the step (4), all the original films embedded with the special marks are guided into image splicing software or algorithm to be spliced to generate an orthoimage with longitude and latitude coordinates, and each pixel point in the orthoimage corresponds to one longitude and latitude coordinate.
7. The method for identifying and positioning the pine wood nematode disease trees based on the original aerial photograph of the unmanned aerial vehicle according to claim 1, wherein the special mark training samples in the step (5) are obtained by cutting the obtained orthographic images into small images with 768 × 768 pixels, and randomly selecting a plurality of small images containing special marks as the special mark training samples, wherein the image marking tool is a labelImg marking tool.
8. The method for identifying and positioning the pine wood nematode disease trees based on the original aerial photograph of the unmanned aerial vehicle as claimed in claim 1, wherein the special mark identification model in the step (6) selects a YOLOv4 model as the special mark identification model, and the value of the last full connection layer of the special mark identification model is set to 1.
9. The pine wood nematode disease tree recognition and positioning method based on the original aerial photograph of the unmanned aerial vehicle as claimed in claim 1, wherein the method for recognizing all the orthographic images in step (7) by special marks comprises the steps of inputting the segmented orthographic image small images into a trained special mark recognition model for recognition, outputting the special mark recognition model as pixel positions and confidence degrees of the special marks in the small images, namely pixel positions of the tree in each small image, setting the confidence degree threshold value to be 0.8, and finally keeping the recognition result of the special marks with the confidence degrees exceeding the threshold value.
10. The method for identifying and positioning the pine wood nematode disease trees based on the original aerial photograph of the unmanned aerial vehicle according to claim 1, wherein the method for converting the original aerial photograph into the longitude and latitude coordinates in the step (8) comprises the steps of obtaining the pixel positions of the central points of all the special marks in the orthographic images according to the pixel positions of the small images in the orthographic images and the pixel positions of the special marks in each small image, and obtaining the longitude and latitude coordinates of the central points according to the corresponding relation between the pixel positions and the longitude and latitude coordinates.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110820396.4A CN113537089A (en) | 2021-07-20 | 2021-07-20 | Pine wood nematode disease infected wood identification and positioning method based on unmanned aerial vehicle aerial photography original sheet |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110820396.4A CN113537089A (en) | 2021-07-20 | 2021-07-20 | Pine wood nematode disease infected wood identification and positioning method based on unmanned aerial vehicle aerial photography original sheet |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113537089A true CN113537089A (en) | 2021-10-22 |
Family
ID=78100488
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110820396.4A Pending CN113537089A (en) | 2021-07-20 | 2021-07-20 | Pine wood nematode disease infected wood identification and positioning method based on unmanned aerial vehicle aerial photography original sheet |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113537089A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116912476A (en) * | 2023-07-05 | 2023-10-20 | 农芯(南京)智慧农业研究院有限公司 | Remote sensing monitoring rapid positioning method and related device for pine wood nematode disease unmanned aerial vehicle |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020122564A1 (en) * | 2001-03-05 | 2002-09-05 | Rhoads Geoffrey B. | Using embedded identifiers with images |
| JP2005037191A (en) * | 2003-07-17 | 2005-02-10 | Asia Air Survey Co Ltd | Automatic orientation method by special mark |
| CN103366555A (en) * | 2013-07-01 | 2013-10-23 | 中国人民解放军第三军医大学第三附属医院 | Aerial image-based traffic accident scene image rapid generation method and system |
| CN108037770A (en) * | 2017-11-22 | 2018-05-15 | 国网山东省电力公司济宁供电公司 | Unmanned plane power transmission line polling system and method based on artificial intelligence |
| CN108710875A (en) * | 2018-09-11 | 2018-10-26 | 湖南鲲鹏智汇无人机技术有限公司 | A kind of take photo by plane road vehicle method of counting and device based on deep learning |
| CN109766830A (en) * | 2019-01-09 | 2019-05-17 | 深圳市芯鹏智能信息有限公司 | A kind of ship seakeeping system and method based on artificial intelligence image procossing |
| CN109948563A (en) * | 2019-03-22 | 2019-06-28 | 华南农业大学 | A kind of withered tree detection localization method of the pine nematode based on deep learning |
| CN110595476A (en) * | 2019-08-30 | 2019-12-20 | 天津航天中为数据***科技有限公司 | Unmanned aerial vehicle landing navigation method and device based on GPS and image visual fusion |
| CN112001966A (en) * | 2020-08-03 | 2020-11-27 | 南京理工大学 | Method for positioning and tracking display screen in flight training AR system |
-
2021
- 2021-07-20 CN CN202110820396.4A patent/CN113537089A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020122564A1 (en) * | 2001-03-05 | 2002-09-05 | Rhoads Geoffrey B. | Using embedded identifiers with images |
| JP2005037191A (en) * | 2003-07-17 | 2005-02-10 | Asia Air Survey Co Ltd | Automatic orientation method by special mark |
| CN103366555A (en) * | 2013-07-01 | 2013-10-23 | 中国人民解放军第三军医大学第三附属医院 | Aerial image-based traffic accident scene image rapid generation method and system |
| CN108037770A (en) * | 2017-11-22 | 2018-05-15 | 国网山东省电力公司济宁供电公司 | Unmanned plane power transmission line polling system and method based on artificial intelligence |
| CN108710875A (en) * | 2018-09-11 | 2018-10-26 | 湖南鲲鹏智汇无人机技术有限公司 | A kind of take photo by plane road vehicle method of counting and device based on deep learning |
| CN109766830A (en) * | 2019-01-09 | 2019-05-17 | 深圳市芯鹏智能信息有限公司 | A kind of ship seakeeping system and method based on artificial intelligence image procossing |
| CN109948563A (en) * | 2019-03-22 | 2019-06-28 | 华南农业大学 | A kind of withered tree detection localization method of the pine nematode based on deep learning |
| CN110595476A (en) * | 2019-08-30 | 2019-12-20 | 天津航天中为数据***科技有限公司 | Unmanned aerial vehicle landing navigation method and device based on GPS and image visual fusion |
| CN112001966A (en) * | 2020-08-03 | 2020-11-27 | 南京理工大学 | Method for positioning and tracking display screen in flight training AR system |
Non-Patent Citations (4)
| Title |
|---|
| RAZAKARIVONY, SEBASTIEN等: "Vehicle detection in aerial imagery : A small target detection benchmark", 《JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION》, vol. 34, 18 November 2015 (2015-11-18), pages 187 - 203, XP029359717, DOI: 10.1016/j.jvcir.2015.11.002 * |
| 赵汉钦: "基于高分辨率影像的单木定位技术研究", 《中国优秀硕士学位论文全文数据库：农业科技辑》, no. 4, 15 April 2020 (2020-04-15), pages 1 - 60 * |
| 金伦等: "基于深度学习的青菜病害区域图像语义分割与定位", 《安徽农业科学》, vol. 48, no. 18, 18 September 2020 (2020-09-18), pages 235 - 238 * |
| 黄华毅等: "Fast R-CNN深度学习和无人机遥感相结合在松材线虫病监测中的初步应用研究", 《环境昆虫学报》, vol. 43, no. 5, 1 July 2021 (2021-07-01), pages 1295 - 1303 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116912476A (en) * | 2023-07-05 | 2023-10-20 | 农芯(南京)智慧农业研究院有限公司 | Remote sensing monitoring rapid positioning method and related device for pine wood nematode disease unmanned aerial vehicle |
| CN116912476B (en) * | 2023-07-05 | 2024-05-31 | 农芯(南京)智慧农业研究院有限公司 | Remote sensing monitoring rapid positioning method and related device for pine wood nematode disease unmanned aerial vehicle |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Underwood et al. | Lidar‐based tree recognition and platform localization in orchards | |
| CN111213155A (en) | Image processing method, device, movable platform, unmanned aerial vehicle and storage medium | |
| AU2020358453B2 (en) | System and method for tracking logs in a wood processing chain | |
| CN109117811A (en) | A kind of system and method based on low-altitude remote sensing measuring technique estimation urban vegetation coverage rate | |
| CN109948563A (en) | A kind of withered tree detection localization method of the pine nematode based on deep learning | |
| CN109815300B (en) | Vehicle positioning method | |
| WO2017194399A1 (en) | Recognition of weed in a natural environment | |
| CN105698715B (en) | A kind of measurement method of the single beam Land area measure system based on unmanned plane | |
| CN111753646A (en) | Agricultural pest detection and classification method fusing population season growth and elimination information | |
| CN112541395A (en) | Target detection and tracking method and device, storage medium and electronic device | |
| CN113255578B (en) | Traffic identification recognition method and device, electronic equipment and storage medium | |
| JP2018100930A (en) | Position specification device, position specification method, and program | |
| US11328388B2 (en) | Image processing apparatus, image processing method, and program | |
| CN113537089A (en) | Pine wood nematode disease infected wood identification and positioning method based on unmanned aerial vehicle aerial photography original sheet | |
| KR20180086745A (en) | Apparatus for processing plant images and method thereof | |
| CN109657540B (en) | Withered tree positioning method and system | |
| CN111552762A (en) | Orchard planting digital map management method and system based on fruit tree coding | |
| WO2021051268A1 (en) | Machine vision-based tree type identification method and apparatus | |
| CN111950524B (en) | Orchard local sparse mapping method and system based on binocular vision and RTK | |
| CN116721344A (en) | Vegetation detection method, device and equipment based on aerial photographing equipment | |
| CN112580489A (en) | Traffic light detection method and device, electronic equipment and storage medium | |
| CN116030324A (en) | Target detection method based on fusion of spectral features and spatial features | |
| CN115294472A (en) | Fruit yield estimation method, model training method, equipment and storage medium | |
| CN113112590B (en) | Ecological variation and vegetation index obtaining method in ecological water delivery engineering | |
| CN114140428A (en) | Method and system for detecting and identifying larch caterpillars based on YOLOv5 |
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 |