CN112101159B - Multi-temporal forest remote sensing image change monitoring method - Google Patents

Multi-temporal forest remote sensing image change monitoring method Download PDF

Info

Publication number
CN112101159B
CN112101159B CN202010921885.4A CN202010921885A CN112101159B CN 112101159 B CN112101159 B CN 112101159B CN 202010921885 A CN202010921885 A CN 202010921885A CN 112101159 B CN112101159 B CN 112101159B
Authority
CN
China
Prior art keywords
image
land
forest
value
current
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.)
Active
Application number
CN202010921885.4A
Other languages
Chinese (zh)
Other versions
CN112101159A (en
Inventor
郭晓妮
董雅雯
杨宁
曾晖
姜灿荣
肖微
付达夫
丁山
周全
胥东海
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Central South Investigation Planning And Design Institute Of State Forestry And Grassland Administration
Original Assignee
Central South Investigation Planning And Design Institute Of State Forestry And Grassland Administration
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Central South Investigation Planning And Design Institute Of State Forestry And Grassland Administration filed Critical Central South Investigation Planning And Design Institute Of State Forestry And Grassland Administration
Priority to CN202010921885.4A priority Critical patent/CN112101159B/en
Publication of CN112101159A publication Critical patent/CN112101159A/en
Application granted granted Critical
Publication of CN112101159B publication Critical patent/CN112101159B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/188—Vegetation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Biology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a multi-temporal forest remote sensing image change monitoring method, which comprises the following steps: monitoring a regional image conversion gray-scale map; calculating a threshold value of the gray value of the target land class; normalizing the remote sensing image according to the threshold value of the gray value of the target land class to obtain normalized front and rear images; respectively carrying out difference value calculation on the normalized front-stage image and the normalized rear-stage image corresponding to different target land types to obtain an image change diagram; respectively carrying out mode filtering and intersection point statistics on the image change graphs of different target land types, then carrying out result integration to obtain a vector graph of preliminary change detection, and after removing broken patches, classifying and screening through a deep neural network model to obtain a vector graph of a final change detection result; and analyzing the change reason of the vector diagram of the final change detection result. According to the invention, the terrain of the monitored area is accurately divided and remote sensing change detection is carried out according to the annual updating result of the forest resource management, so that an accurate change detection result is obtained.

Description

Multi-temporal forestry remote sensing image change monitoring method
Technical Field
The invention relates to the field of remote sensing change monitoring, in particular to a multi-temporal forest remote sensing image change monitoring method.
Background
The remote sensing change monitoring is to utilize multi-temporal remote sensing images, extract change information by adopting various image recognition methods, quantitatively analyze and determine the characteristics and the process of surface change. It relates to the type, distribution and variation of the change, i.e. the ground type, boundary line and variation trend before and after the change need to be determined, and then the characteristics and forming reasons of the dynamic changes are analyzed. The remote sensing change monitoring is divided into three methods of pixel level, characteristic level and target level according to the level of a processing object. The research on the detection of the characteristic level and target level changes is not mature, and the detection is difficult to be used in forestry practice in scale.
Pixel level change detection includes difference image methods and other pixel change detection methods, which are generally used because of their low universality compared to difference image methods. The difference image method comprises the following steps:
the thresholding method selects a threshold value to carry out thresholding on the difference image so as to distinguish the pixels with variation type and pixels with non-variation type, and has the advantages that the method is simple and clear, and has the defects that the context information of the image is not fully utilized and is easily influenced by factors such as sensor noise, registration error and the like;
The mode classification method utilizes a classifier to classify different sample data sets so as to obtain a change detection result, and has the advantages of overcoming the inaccuracy of a simple thresholding method, having the defects of incapability of utilizing spatial context information, and providing supervision information by manual intervention of methods of an artificial neural network and a support vector machine;
the multi-scale analysis method utilizes the multi-scale geometric analysis method to carry out change detection, thereby overcoming the influence of factors such as sensor noise, registration error and the like, having the advantages of overcoming the factors such as the sensor noise, the registration error and the like, and having the defects of detail loss, increased calculation time and unsolved multi-scale result processing technical difficulties;
the Markov random field method utilizes a Markov random field model to simulate space context information so as to obtain a change detection result, has the advantages of overcoming irrationality of parameterization probability model assumption of a difference image, and utilizes the context information, and has the defects of large calculation amount and long calculation time aiming at high-resolution images.
Dividing forest lands into two categories of forest lands and non-forest lands according to a land classification standard of forest resource second-class investigation; wherein, the woodland comprises woodland (0110), thinning land (0120), shrub woodland (0130), non-woodland (0140), nursery land (0150), non-woodland (0160), Yilin woodland (0170) and auxiliary forestry production land (0180); the non-forestry land includes cultivated land (0210), grassland (0220), water area (0230), unused land (0240) and construction land (0250). The annual updating of the forest resource management is carried out every year on the basis of the second-class investigation of the forest resources, so that the actual condition that the forest resources planted in the previous year are covered is reflected timely and accurately.
According to the land classification standard of the second-class survey of the forest resources, the remote sensing change monitoring result is processed by combining the actual planting and covering condition reflected by the annual updating result of the forest resource management in one picture so as to obtain a more accurate result, and the method is worthy of research and discussion.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: aiming at the technical problems in the prior art, the invention provides a multi-temporal forest remote sensing image change monitoring method, which is used for accurately dividing the terrain of a monitoring area according to the annual updating result of forest resource management, and carrying out remote sensing change monitoring to obtain an accurate monitoring result.
In order to solve the technical problems, the technical scheme provided by the invention is as follows:
a multi-temporal forestry remote sensing image change monitoring method comprises the following steps:
s1) monitoring the regional image transformation gray-scale map: respectively converting red light wave bands of front and rear two-stage images of a monitoring area into gray level images;
s2) calculating a threshold value of the gradation value of the target land class: analyzing the magnitude relation of the gray values of the red light wave bands corresponding to different land types according to the land type division standard of the second type investigation of the forest resources and the annual updating result of the forest resource management of one image, and then calculating the threshold value of the gray value corresponding to the target land type according to the change of the target land type in the later image, wherein the target land type comprises a covered forest land and a non-covered forest land;
S3) remote sensing image normalization: processing the front and rear images according to the gray value threshold values corresponding to the covered forest lands and the non-covered forest lands respectively to obtain the processed front and rear images corresponding to the covered forest lands and the non-covered forest lands, and then normalizing the processed front and rear images to obtain a normalized front image A1 and a normalized rear image A2 corresponding to the covered forest lands and a normalized front image A1 'and a normalized rear image A2' corresponding to the non-covered forest lands;
s4) normalized result difference calculation: calculating difference values of corresponding pixels in the normalized front-stage image A1 and the normalized rear-stage image A2 corresponding to the forest land with vegetation coverage to obtain a corresponding image change map B, and calculating difference values of corresponding pixels in the normalized front-stage image A1 ' and the normalized rear-stage image A2 ' corresponding to the forest land without vegetation coverage to obtain a corresponding image change map B ';
s5) mode filtering: respectively restoring the graph cavity phenomena of the image change diagrams B and B 'by using mode filtering to obtain restored image change diagrams B1 and B1' corresponding to forest lands with vegetation coverage and forest lands without vegetation coverage;
S6) focus statistics: respectively utilizing focus statistics to combine the adjacent plaque breaking phenomena of the plaques in the repaired image change images B1 and B1 'to obtain secondary repaired image change images B2 and B2' corresponding to forest lands with vegetation coverage and forest lands without vegetation coverage;
s7) result integration: sequentially carrying out binarization and vectorization on the image change images B2 and B2' subjected to secondary restoration, and merging to obtain a vector image D for primary change detection;
s8) removing broken plaques: calculating the actual area of the vector diagram D of the preliminary change detection, and removing the broken patches with the actual area smaller than a preset value in the diagram;
s9) classification and screening of the deep neural network model: inputting the vector diagram D of the preliminary change detection after the broken patches are removed into a deep neural network model, taking the vector diagram D of the preliminary change detection as a further judgment object range, classifying and screening through the neural network model, and removing unchanged patches caused by image radiation errors in the diagram to obtain a vector diagram D1 of the final change detection;
s10) analyzing the cause of the change: and extracting the average gray value of the image spot in the vector image D1 of the final change detection, and analyzing the change reason according to the gray value.
Further, the step of calculating the threshold of the grayscale value corresponding to the target land type in step S2) specifically includes:
S21) selecting covered woodlands or uncovered woodlands as current land types, selecting early-stage images or late-stage images as current images, finding areas where reference land types corresponding to the current land types are located in the current images, converting surface data in the areas where the reference land types are located into point data, wherein the reference land types corresponding to the covered woodlands are cultivated lands, and the reference land types corresponding to the uncovered woodlands are construction lands or unused lands;
s22), acquiring gray values of all point data, firstly carrying out abnormal value detection through boxplot analysis, removing abnormal values, then carrying out K-means clustering to obtain at least one clustering center, and selecting the smallest clustering center as a threshold value of the gray value corresponding to the current land type in the current image.
Further, step S3) specifically includes the following steps:
s31) selecting forest lands with plants covered or forest lands without plants covered as the current land class, and determining the gray values of all pixels in the processed later-period image corresponding to the current land class according to the threshold value of the gray value corresponding to the current land class in the later-period image, wherein the function expression is as follows:
Figure BDA0002667001730000031
in the above formula, xiIs the gray value E of the pixel i in the processed later-stage image corresponding to the current land class 1A threshold value of a gray value corresponding to the current land class in the later-stage image;
s32) determining the gray values of all pixels in the processed early-stage image corresponding to the current land type according to the threshold of the gray value corresponding to the current land type in the early-stage image, wherein the function expression is as follows:
Figure BDA0002667001730000032
in the above formula, yiIs the gray value E of the pixel i in the processed early-stage image corresponding to the current land type2A threshold value of a gray value corresponding to the current land class in the previous image;
s33) normalizing the gray values of the pixels in the processed front and rear images corresponding to the current land type to obtain the normalized front and rear images corresponding to the current land type, wherein the function expression is as follows:
Figure BDA0002667001730000041
in the above formula, X'iIs the gray value, Y ', of the pixel i in the normalized later-stage image corresponding to the current ground class'iIs the gray value, x, of the pixel i in the normalized previous image corresponding to the current land typeiIs the gray value y of the pixel i in the processed later-stage image corresponding to the current land classiIs the gray value, x, of the pixel i in the processed earlier-stage image corresponding to the current land typeimaxThe maximum value, x, of the gray level of each pixel in the processed later-stage image corresponding to the current terrainiminIs the minimum value, y, of the gray level of each pixel in the processed later-stage image corresponding to the current terrain imaxThe maximum value y of the gray level of each pixel in the processed early-stage image corresponding to the current land typeiminThe minimum value of the gray scale of each pixel in the processed early-stage image corresponding to the current land type is obtained.
Further, in step S4), the function expression of calculating the difference value of the corresponding pixels in the normalized previous and subsequent images one by one is as follows:
Figure BDA0002667001730000042
in the above formula, X'iIs the gray value, Y ', of the pixel i in the normalized later-stage image'iIs the gray value of the pixel i in the normalized previous image, and e is the preset difference threshold.
Further, the specific step of the mode filtering in step S5) includes: and aiming at the image change image B or B', selecting one pixel in the image change image as a current pixel, then setting a rectangular window with the current pixel as a center and taking one pixel as a distance, and if more than five eighths of adjacent pixels in other pixels around the current pixel in the rectangular window have the same value, replacing the value of the current pixel with the value of the adjacent pixels.
Further, the specific step of the focus statistics in step S6) includes: and aiming at the repaired image change image B1 or B1', selecting one pixel in the repaired image change image as a current pixel, then setting a rectangular window with two pixels as a distance by taking the current pixel as a center, counting the average value of the values of all the pixels in the rectangular window, and taking the average value of the values of all the pixels as the value of the current pixel.
Further, step S7) specifically includes the following steps:
s71) binarizing the image change maps B and B 'respectively to obtain binary maps C0 and C0' of preliminary change monitoring results corresponding to forest lands with plant coverage and forest lands without plant coverage;
s72) respectively vectorizing the binary images C0 and C0 'to obtain vector images D0 and D0';
s73) merging the vectorized vector images D0 and D0' to obtain a vector image D for preliminary change detection.
Further, step S9) specifically includes the following steps:
s91) inputting the vector diagram D of the preliminary change detection after the broken plaque is removed into a deep neural network model;
s92) taking the vector diagram D of the preliminary change detection as a further judgment object range, classifying and screening through a neural network model, and then removing unchanged image spots caused by image radiation errors in the diagram to obtain the vector diagram D1 of the final change detection.
Further, step S10) specifically includes the following steps:
s101) converting all surface data of the pattern spots in the vector diagram D1 of the final change detection into point data;
s102) extracting the gray value of the corresponding pixel according to the point data;
s103) abnormal value detection is carried out on the gray values of all the extracted pixels through boxline graph analysis, K-means clustering is carried out after the abnormal values are removed, the gray value of the minimum clustering center is used as the gray value after the clustering analysis, and the change reason is analyzed according to the size relation between the gray value after the clustering analysis and the gray value of the forest lands planted and uncovered.
The invention further provides a computer-readable storage medium, which stores a computer program programmed or configured to implement the multi-temporal forest remote sensing image change monitoring method.
Compared with the prior art, the invention has the advantages that:
(1) according to the method, the land classification standard of the second-class investigation of the forest resources is combined with the annual updating result of the forest resource management, the land classification of a monitoring area is divided into covered forest lands and uncovered forest lands, and then change detection is respectively carried out;
(2) according to the method, a threshold value setting normalization method is adopted to obtain a preliminary change detection result, a machine learning method is introduced, and a deep neural network model is specifically used for classification and screening, so that firstly, the influence caused by different periods of remote sensing image difference is effectively reduced, and the universality is improved; secondly, errors of the model caused by artificial factors in the sample selection process are avoided, and the precision is improved; thirdly, the operation time is reduced, and the efficiency is improved; fourthly, the mandatory requirements of the invention on hardware conditions are reduced, and the popularization is facilitated;
(3) The invention introduces a method of expansion in image morphology, utilizes modes of mode filtering and focus statistics, and effectively avoids the phenomena of adjacent broken patches of patches and abnormal pattern holes in an image change image through the thought of a matrix.
Drawings
FIG. 1 is a schematic step diagram of an embodiment of the present invention.
FIG. 2 is a detailed flow chart of an embodiment of the present invention.
Fig. 3 is a representation of mode filtering.
Fig. 4 is a focus statistics rendering.
Detailed Description
The invention is further described below with reference to the drawings and specific preferred embodiments of the description, without thereby limiting the scope of protection of the invention.
As shown in fig. 1, the multi-temporal forest remote sensing image change monitoring method of the invention comprises the following steps:
s1) converting the monitoring area image into a gray-scale image: in order to facilitate the use of basic forestry departments, high-resolution images issued by national forestry and grassland offices are preprocessed false color images and are respectively formed by fusing three wave bands of infrared, red and green, wherein the red light wave band is most suitable for forestry remote sensing image change detection, so that the red light wave bands of the front and rear two-stage images of a monitoring area are respectively converted into gray maps in the embodiment;
S2) calculating a threshold value of the gradation value of the target land class: the land classification standard of the second type of forest resource investigation divides the land into two main types of forest land and non-forest land, for a red light wave band, the higher the gray value is, the lower the vegetation coverage rate of the land, the prior art does not distinguish forest lands with covered plants from forest lands without covered plants, and according to the annual updating result of forest resource management, we find that:
for non-forest lands, the gray values of the red light wave bands corresponding to different land types are sorted from large to small into construction land, unused land, grassland, cultivated land and water areas;
taking the woodland and the non-woodland into consideration integrally, and sequencing the gray values of the red light wave band corresponding to the land types from large to small into a construction land, an unused land, a woodland without planting coverage, a grassland, a cultivated land and a woodland with planting coverage;
in addition, according to the actual situation of land type change and the difference of gray values of red light wave band, the covered forest lands can be changed into construction lands, unused lands, uncovered forest lands, grasslands and cultivated lands in the later-stage image, and the uncovered forest lands can be changed into construction lands and unused lands in the later-stage image;
therefore, in this embodiment, according to the land classification standard of the second class investigation of forest resources and the updated data of the forest resource management "one map" year end, the size relationship of the gray values of the red light bands corresponding to different land classes is analyzed, then the threshold of the gray values corresponding to the target land class is calculated according to the land class change in the later-stage image of the target land class, the forest change detection mainly aims at the condition that the forest land is changed into a non-forest land or a non-covered forest land, the target land class of this embodiment includes covered forest lands and uncovered forest lands, for the forest land with covered vegetation, the forest land with covered vegetation can be converted into the non-forest land or the uncovered forest land, according to the size relationship of the gray values of the red light bands corresponding to different land classes, the threshold of the gray values corresponding to the covered forest land is the gray value after the area cluster analysis of the land class being cultivated land, and the gray value of the pixel of the area with covered forest land in the later-stage image after the conversion should be greater than the area in the image with the land class being cultivated land class being the gray value Gray values after area cluster analysis of cultivated lands (namely, forest lands with covered plants in the later-stage image are converted into non-forest lands or non-covered forest lands), gray values of pixels of areas with covered forest lands in the earlier-stage image are smaller than gray values after area cluster analysis, in which the land type in the image is cultivated lands (namely, forest lands with covered plants in the earlier-stage image); for the non-planted and covered woodland, the non-planted and covered woodland can be converted into a construction land or an unused land, according to the magnitude relation of the gray values of the red light wave bands corresponding to different land types, the threshold value of the gray value corresponding to the non-planted and covered woodland is the gray value after the region cluster analysis of the land type as the construction land or the unused land, the gray value of the pixel of the region after the conversion of the non-planted and covered woodland in the later-stage image is greater than the gray value after the region cluster analysis of the land type as the construction land or the unused land in the image (namely, the non-planted and covered woodland in the later-stage image is converted into the non-forestry land), and the gray value of the pixel of the region where the non-planted and covered woodland in the earlier-stage image is smaller than the gray value after the region cluster analysis of the land type as the construction land or the unused land in the image (namely, the non-planted and covered woodland in the earlier-stage image);
According to the above content, the step of obtaining the threshold of the gray-level value corresponding to the target land category in this embodiment includes:
s21) selecting covered forest lands or non-covered forest lands as current land types, selecting early-stage or late-stage images as current images, finding out the area where the reference land type corresponding to the current land type is located in the current images, converting the surface data in the area where the reference land type is located into point data, wherein the reference land type corresponding to the covered forest lands is cultivated land, and the reference land type corresponding to the non-covered forest lands is construction land or unused land;
s22) obtaining gray values of all point data, firstly carrying out abnormal value detection through boxplot analysis, removing abnormal values, then carrying out K-means clustering to obtain at least one clustering center, and selecting the smallest clustering center as a threshold value of the gray value corresponding to the current land in the current image;
obtaining the threshold of the gray value corresponding to the forest land planted and covered in the early image, the threshold of the gray value corresponding to the forest land not planted and covered in the early image, the threshold of the gray value corresponding to the forest land planted and covered in the later image and the threshold of the gray value corresponding to the forest land not planted and covered in the later image according to the steps S21) to S22);
S3) remote sensing image normalization: according to the analysis content in step S2), in this embodiment, the threshold of the gray value corresponding to the target land is used as the boundary value between the forest land and the non-forest land corresponding to the target land, and meanwhile, the threshold of the gray value in this embodiment is the gray value of the minimum cluster center after the abnormal value is removed from the region where the reference land corresponding to the target land is located, so that the intersection of data between the forest land and the non-forest land can be ignored, as shown in fig. 2, after the threshold of the gray value corresponding to the forest land with vegetation coverage and the forest land without vegetation coverage is obtained, the ranges of the processed front and rear images corresponding to the forest land with vegetation coverage and the forest land without vegetation coverage are selected according to the threshold of the gray value corresponding to the forest land with vegetation coverage and the forest land without vegetation coverage, so that the processed front image does not contain non-forest land, and the processed rear image does not contain forest land, thereby reducing the post-calculation amount, and then normalizing the processed early-stage image and the processed post-stage image to obtain a normalized early-stage image A1 and a normalized post-stage image A2 corresponding to the forest land with planting coverage, and a normalized early-stage image A1 'and a normalized post-stage image A2' corresponding to the forest land without planting coverage;
S4) normalized result difference calculation: as shown in fig. 2, for forest lands with covered vegetation, performing difference calculation on the corresponding normalized front-stage image a1 and the corresponding normalized back-stage image a2 according to the normalized front-stage image and the normalized back-stage image and corresponding pixels in the normalized back-stage image and the normalized back-stage image, one by one, to obtain a corresponding image change map B, and for forest lands without covered vegetation, performing difference calculation on the corresponding normalized front-stage image a1 'and the normalized back-stage image a 2' according to corresponding pixels in the normalized front-stage image and the normalized back-stage image, one by one, to obtain a corresponding image change map B ', where the image change maps B and B' are change detection results of the forest lands with covered vegetation and the forest lands without covered vegetation;
s5) mode filtering: as shown in fig. 2, the pattern hole phenomenon is restored by using mode filtering for the image change map B corresponding to the forest land with vegetation coverage and the image change map B 'corresponding to the forest land without vegetation coverage, so as to obtain a restored image change map B1 corresponding to the forest land with vegetation coverage and a restored image change map B1' corresponding to the forest land without vegetation coverage;
s6) focus statistics: as shown in fig. 2, for the repaired image change map B1 corresponding to the forest land with vegetation coverage and the repaired image change map B1 'corresponding to the forest land without vegetation coverage, a secondary repaired image change map B2 corresponding to the forest land with vegetation coverage and a secondary repaired image change map B2' corresponding to the forest land without vegetation coverage are obtained by respectively using the near-plaque phenomenon in the focus statistics merged map;
S7) result integration: as shown in fig. 2, binarizing the image change map B2 after the secondary restoration corresponding to the forest land with covered planting and the image change map B2' after the secondary restoration corresponding to the forest land without covered planting, and vectorizing the result after binarization, wherein the image spots in the vectorized result are the changed areas of the forest land with covered planting and the forest land without covered planting after the graphic restoration, and then combining the vectorized result to obtain a vector map D for primary change detection, wherein the image spots in the vector map D for primary change detection are the areas of the forest land which is obtained primarily in this embodiment and is changed into the non-forest land or the forest land without covered planting;
s8) removing broken plaques: as shown in fig. 2, calculating the actual area of the vector diagram D of the preliminary change detection, and removing the broken patches with the actual area smaller than the preset value;
s9) classification and screening of the deep neural network model: as shown in fig. 2, the vector diagram D of the preliminary change detection after the removal of the broken patches is input into a deep neural network model, the vector diagram D of the preliminary change detection is taken as a further judgment object range, and after classification and screening are performed by the neural network model, the unchanged patches caused by the image radiation errors in the diagram are removed, so as to obtain a vector diagram D1 of the final change detection, wherein the patches in the vector diagram D1 of the final change detection are the accurate regions of the forest land obtained in this embodiment, which is changed into a non-forest land or a forest land without planting coverage;
S10) analyzing the cause of change: as shown in fig. 2, the gradation value of the patch in the vector image D1 of the final change detection is extracted, and the cause of change is analyzed based on the gradation value.
Step S3) of this embodiment specifically includes the following steps:
s31) selecting forest lands with plants covered or forest lands without plants covered as the current land class, and determining the gray values of all pixels in the processed later-period image corresponding to the current land class according to the threshold value of the gray value corresponding to the current land class in the later-period image, wherein the function expression is as follows:
Figure BDA0002667001730000081
in the above formula, xiIs the gray value E of the pixel i in the processed later-stage image corresponding to the current land class1A threshold value of a gray value corresponding to the current land class in the later-stage image;
s32) determining the gray values of all pixels in the processed early-stage image corresponding to the current land type according to the threshold of the gray value corresponding to the current land type in the early-stage image, wherein the function expression is as follows:
Figure BDA0002667001730000082
in the above formula, yiIs the gray value E of the pixel i in the processed early-stage image corresponding to the current land type2A threshold value of a gray value corresponding to the current land class in the previous image;
s33) normalizing the gray values of the pixels in the processed front and rear images corresponding to the current land type to obtain the normalized front and rear images corresponding to the current land type, wherein the function expression is as follows:
Figure BDA0002667001730000091
X 'in the above formula'iIs the gray value Y 'of the pixel i in the normalized later-period image corresponding to the current place'iIs the gray value, x, of the pixel i in the normalized previous image corresponding to the current land typeiIs gray of pixel i in the processed later-stage image corresponding to the current place classValue of yiIs the gray value, x, of the pixel i in the processed earlier-stage image corresponding to the current land typeimaxThe maximum value, x, of the gray scale of each pixel in the processed later-period image corresponding to the current place classiminIs the minimum value, y, of the gray level of each pixel in the processed later-stage image corresponding to the current terrainimaxIs the maximum value y of the gray scale of each pixel in the processed early-stage image corresponding to the current land classiminThe minimum value of the gray scale of each pixel in the processed early-stage image corresponding to the current land type.
In this embodiment, the normalized front and rear images corresponding to the covered woodland and the normalized front and rear images corresponding to the uncovered woodland can be obtained according to steps S31) to S33):
in step S31), if the current land type is a forest land with covered vegetation, the threshold of the gray value corresponding to the current land type in the later image as the boundary value between the forest land and the non-forest land is the threshold of the gray value corresponding to the forest land with covered vegetation. Thus E 1Eliminating abnormal values from the gray values of all point data of the cultivated land area in the later-stage image to obtain the gray value of the minimum clustering center, obtaining a processed later-stage image A01 corresponding to the forest land with planting coverage according to the formula (1), and if the current land type is the forest land without planting coverage, taking the gray value threshold corresponding to the current land type in the later-stage image of the boundary value of the forest land and the non-forest land as the gray value threshold corresponding to the forest land without planting coverage, so E1And (3) eliminating abnormal values from the gray values of all point data of the areas of the construction land or the unused land in the later-stage image to obtain the gray value of the minimum clustering center, and obtaining a processed later-stage image A01' corresponding to the woodland without being covered by plants according to the formula (1).
Step S32), if the current land type is a covered forest land, the threshold of the gray value corresponding to the current land type in the previous image as the boundary value between the forest land and the non-forest land is the threshold of the gray value corresponding to the covered forest land, therefore E2Eliminating abnormal values for the gray values of all point data of the cultivated land area in the earlier-stage image, then obtaining the gray value of the minimum clustering center, and obtaining the corresponding forest land covered with plants according to the formula (2)If the current land type of the processed early-stage image a02 is a non-planted forest land, the threshold of the gray value corresponding to the current land type in the early-stage image as the boundary value between the forest land and the non-forest land is the threshold of the gray value corresponding to the non-planted forest land, therefore E 2The gray value of the minimum cluster center is obtained after the abnormal value is removed from the gray values of all the point data of the areas of the construction land or the unused land in the prior image, and the processed prior image A02' corresponding to the forest land without vegetation coverage is obtained according to the formula (2).
In step S33), if the current land type is a forest land with covered vegetation, the gray values of the pixels in the corresponding processed early-stage image a02 and the processed late-stage image a01 are normalized according to formula (3), so that the corresponding normalized early-stage image a1 and the normalized late-stage image a2 are obtained, and if the current land type is a forest land without covered vegetation, the gray values of the pixels in the corresponding processed early-stage image a02 'and the processed late-stage image a 01' are normalized according to formula (3), so that the corresponding normalized early-stage image a1 'and the normalized late-stage image a 2' are obtained.
In step S4) of this embodiment, the functional expression of calculating the difference value of the corresponding pixels in the normalized first and second-stage images one by one is as follows:
Figure BDA0002667001730000101
in the above formula, X'i-Y'iIs the value of pixel i in the image variation graph, X'iIs the gray value, Y ', of the pixel i in the normalized later-stage image'iThe gray value of the pixel i in the normalized previous image is, e is a preset difference threshold, the difference threshold is mainly determined according to a test result, and the landform is also subdivided according to the actual situation of each monitoring area.
According to the formula (4) in the step S4), for forest lands with covered vegetation, performing difference calculation on corresponding pixels in the normalized previous image a1 and the normalized subsequent image a2 one by one to obtain a corresponding image change map B, and for forest lands without covered vegetation, performing difference calculation on corresponding pixels in the normalized previous image a1 'and the normalized subsequent image a 2' one by one to obtain a corresponding image change map B ', where the image change maps B and B' are binary maps.
The present embodiment introduces a method of dilation in image morphology, step S5) restores the phenomenon of abnormal pattern holes by using mode filtering, and step S6) merges the near-fragmented phenomenon of the patch by using focus statistics.
The principle of mode filtering is shown in fig. 3, and eight nearest neighboring pixels (3 × 3 window) with a value of "-3" in the graph are analyzed, and the data values thereof relate to "7", "5" and "4", and the numbers thereof are 5, 2 and 1, respectively. The number of "7" is the largest, the pixels have the same value and are adjacent, and more than five eighths of the connected pixels have the same value, then "-3" is replaced with "7". Therefore, the specific step of the mode filtering in step S5) of this embodiment includes: and aiming at the image change image B or B', selecting one pixel in the image change image as a current pixel, then setting a rectangular window with the current pixel as a center and taking one pixel as a distance, and if more than five eighths of adjacent pixels in other pixels around the current pixel in the rectangular window have the same value, replacing the value of the current pixel with the value of the adjacent pixels.
The principle of focus statistics is shown in fig. 4, in the figure, a current pixel is taken as a center, a 5 × 5 rectangle is taken as a window for analysis, and the average value of the current pixel and the average value of the peripheral 24 pixels are counted as the value of the pixel, for example, the sum of the values of the 5 × 5 window is 125 for 5 in the center position of the graph a, and the statistical result is 5 if the average value of the pixels in the calculation field is 5. The method can meet the requirement of adjacent plaque fragmentation of combined plaque. Therefore, the specific step of the focal point statistics in step S6) of this embodiment includes: and aiming at the repaired image change image B1 or B1', selecting one pixel in the repaired image change image as a current pixel, then setting a rectangular window with two pixels as a distance by taking the current pixel as a center, counting the average value of the values of all the pixels in the rectangular window, and taking the average value of the values of all the pixels as the value of the current pixel.
Step S7) of this embodiment specifically includes the following steps:
s71), because the values of the pixels in the secondary repaired image change maps B2 and B2 ' calculated in the step S6) may have decimal numbers and do not meet the requirement of binary maps, so that the subsequent calculation and analysis need to be performed with binarization again, in this embodiment, the secondary repaired image change maps B and B ' corresponding to forest lands planted and uncovered forest lands are respectively binarized, the difference threshold of binarization is determined mainly according to the test result, and the landform is subdivided according to the actual situation of each monitoring area, in this embodiment, the landform is divided into two categories, namely a plain area and a mountain area, and the difference threshold is respectively set, for the secondary repaired image change maps B and B ', if the value of the pixel in the secondary repaired image change map is less than the difference threshold, the value of the pixel is set to 0, and if the value of the pixel is greater than the difference threshold value, setting the value of the pixel to be 1, and thus obtaining binary images C0 and C0' corresponding to the forest lands with vegetation coverage and the forest lands without vegetation coverage.
S72) respectively vectorizing binary images C0 and C0 'corresponding to forest lands with planting coverage and forest lands without planting coverage to obtain vector images D0 and D0'; the pattern spots in the vector diagram D0 are the changed areas of the covered woodland planted after the graph is repaired, and the pattern spots in the vector diagram D0' are the changed areas of the uncovered woodland planted after the graph is repaired;
s73) merging the vector diagram D0 corresponding to the forest land with planting coverage and the vector diagram D0' corresponding to the forest land without planting coverage after vectorization to obtain a vector diagram D for preliminary change detection.
In step S9) of this embodiment, the classification model is a deep neural network model (DNN), and the training samples are the forest supervision results of the previous year. DNN is one of the most widely used machine learning algorithms, forest remote sensing images at the present stage are all high-resolution images, the time consumption for detecting the change of the images by using a deep neural network method is too long, and the requirement on the configuration of equipment is high. In this embodiment, after preliminary change detection based on a threshold is performed, classification and screening are performed by using DNN, so that manual intervention is reduced by using an existing training sample while inaccuracy of a simple thresholding method is overcome, and step S9) specifically includes the following steps:
S91) inputting the vector diagram D of the preliminary change detection after the broken plaque is removed into a deep neural network model;
s92) taking the vector diagram D of the preliminary change detection as a further judgment object range, classifying and screening through a neural network model, and then removing unchanged image spots caused by image radiation errors in the diagram to obtain the vector diagram D1 of the final change detection.
Step S10) of this embodiment specifically includes the following steps:
s101) converting all surface data of the pattern spots in the vector diagram D1 of the final change detection into point data;
s102) extracting the gray value of the corresponding pixel according to the point data;
s103) abnormal value detection is carried out on the gray values of all the extracted pixels through boxplot analysis, K-means clustering is carried out after the abnormal values are eliminated, the gray value of the minimum clustering center is used as the gray value after clustering analysis, and the change reasons are analyzed according to the size relation between the gray value after clustering analysis and the gray value threshold of the forest land with planting coverage and the forest land without planting coverage.
According to the requirement of annual updating of forest resource management, the change reasons can be divided into land type or forest phase change (40) caused by occupation of forest lands (10), forest felling (20), reclamation (30), disasters and the like according to construction projects, and recognizable land type or forest phase change (50) and other changes (60) caused by afforestation updating. In step S103), if the gray value after the cluster analysis is closer to the gray value after the cluster analysis of the construction land, the change cause is the change of land or forest facies caused by the forest land occupation or disaster of the construction project, and if the gray value after the cluster analysis is closer to the gray value after the cluster analysis of the cultivated land, the change cause is the change of land or forest facies caused by forest felling, reclamation, identifiable forest updating, or other changes, and then the concrete change cause is filled in with suggestions in combination with the information of land which is updated in the "one map" year of forest resource management, for example, the change cause can not be filled in as forest felling except for the areas of arbor forest with the diameter greater than 5cm and which are accumulated.
The invention further provides a computer-readable storage medium, which stores a computer program programmed or configured to implement the multi-temporal forest remote sensing image change monitoring method.
The foregoing is illustrative of the preferred embodiments of the present invention and is not to be construed as limiting the invention in any way. Although the present invention has been described with reference to the preferred embodiments, it is not intended to be limited thereto. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical spirit of the present invention should fall within the protection scope of the technical scheme of the present invention, unless the technical spirit of the present invention departs from the content of the technical scheme of the present invention.

Claims (10)

1. A multi-temporal forestry remote sensing image change monitoring method is characterized by comprising the following steps:
s1) monitoring the regional image transformation gray-scale map: respectively converting red light wave bands of front and rear two-stage images of a monitoring area into gray level images;
s2) calculating a threshold value of the gradation value of the target land class: analyzing the magnitude relation of the gray values of the red light wave bands corresponding to different land types according to the land type division standard of the second type investigation of the forest resources and the annual updating result of the forest resource management of one image, and then calculating the threshold value of the gray value corresponding to the target land type according to the change of the target land type in the later image, wherein the target land type comprises a covered forest land and a non-covered forest land;
S3) remote sensing image normalization: processing the front-stage image and the rear-stage image according to the threshold values of the gray values corresponding to the covered forest land and the non-covered forest land respectively to obtain the processed front-stage image and the processed rear-stage image corresponding to the covered forest land and the non-covered forest land, and then normalizing the processed front-stage image and the processed rear-stage image to obtain a normalized front-stage image A1 and a normalized rear-stage image A2 corresponding to the covered forest land and a normalized front-stage image A1 'and a normalized rear-stage image A2' corresponding to the non-covered forest land;
s4) normalized result difference calculation: calculating difference values of corresponding pixels in the normalized front-stage image A1 and the normalized rear-stage image A2 corresponding to the forest land with vegetation coverage to obtain a corresponding image change diagram B, and calculating difference values of corresponding pixels in the normalized front-stage image A1 ' and the normalized rear-stage image A2 ' corresponding to the forest land without vegetation coverage to obtain a corresponding image change diagram B ';
s5) mode filtering: restoring the graph cavity phenomena of the image change diagrams B and B 'by using mode filtering respectively to obtain restored image change diagrams B1 and B1' corresponding to forest lands with vegetation coverage and forest lands without vegetation coverage;
S6) focus statistics: respectively utilizing focus statistics to combine the adjacent plaque breaking phenomena of the plaques in the repaired image change images B1 and B1 'to obtain secondary repaired image change images B2 and B2' corresponding to forest lands with vegetation coverage and forest lands without vegetation coverage;
s7) result integration: sequentially carrying out binarization and vectorization on the image change images B2 and B2' subjected to secondary restoration, and merging to obtain a vector image D for primary change detection;
s8) removing broken plaques: calculating the actual area of the vector diagram D of the preliminary change detection, and removing the broken patches with the actual area smaller than a preset value in the diagram;
s9) classification and screening of the deep neural network model: inputting the vector diagram D of the preliminary change detection after the broken patches are removed into a deep neural network model, taking the vector diagram D of the preliminary change detection as a further judgment object range, classifying and screening through the neural network model, and removing unchanged patches caused by image radiation errors in the diagram to obtain a vector diagram D1 of the final change detection;
s10) analyzing the cause of the change: and extracting the average gray value of the image spot in the vector image D1 of the final change detection, and analyzing the change reason according to the gray value.
2. The multi-temporal forest remote sensing image change monitoring method according to claim 1, wherein the step of calculating the threshold value of the gray value corresponding to the target land category in step S2) specifically comprises:
S21) selecting covered forest lands or uncovered forest lands as a current land type, selecting an early-stage image or a later-stage image as a current image, finding an area where a reference land type corresponding to the current land type is located in the current image, converting surface data in the area where the reference land type is located into point data, wherein the reference land type corresponding to the covered forest lands is cultivated land, and the reference land type corresponding to the uncovered forest lands is construction land or unused land;
s22) obtaining the gray value of all point data, firstly carrying out abnormal value detection through boxplot analysis, eliminating abnormal values, then carrying out K-means clustering to obtain at least one clustering center, and selecting the smallest clustering center as the threshold value of the gray value corresponding to the current land type in the current image.
3. The multi-temporal forest remote sensing image change monitoring method according to claim 1, wherein step S3) includes the steps of:
s31) selecting forest lands with planting coverage or forest lands without planting coverage as the current land type, and determining the gray values of all pixels in the processed later-period image corresponding to the current land type according to the threshold value of the gray value corresponding to the current land type in the later-period image, wherein the function expression is as follows:
Figure FDA0002667001720000021
In the above formula, xiIs the gray value E of the pixel i in the processed later-stage image corresponding to the current land class1A threshold value of a gray value corresponding to the current land class in the later-stage image;
s32) determining the gray values of all pixels in the processed early-stage image corresponding to the current land type according to the threshold of the gray value corresponding to the current land type in the early-stage image, wherein the function expression is as follows:
Figure FDA0002667001720000022
in the above formula, yiIs the gray value E of the pixel i in the processed early-stage image corresponding to the current land type2A threshold value of a gray value corresponding to the current land class in the previous image;
s33) normalizing the gray values of the pixels in the processed images in the previous and later periods corresponding to the current land type to obtain the normalized images in the previous and later periods corresponding to the current land type, wherein the function expression is as follows:
Figure FDA0002667001720000023
in the above formula, X'iIs the gray value of the pixel i in the normalized later-stage image corresponding to the current land class, Yi' is the gray value, x, of the pixel i in the normalized prior image corresponding to the current land categoryiIs the gray value y of the pixel i in the processed later-stage image corresponding to the current land classiIs the gray value, x, of the pixel i in the processed earlier-stage image corresponding to the current land typeimaxThe maximum value, x, of the gray scale of each pixel in the processed later-period image corresponding to the current place class iminIs the minimum value of the gray level of each pixel in the processed later-period image corresponding to the current land type, yimaxThe maximum value y of the gray level of each pixel in the processed early-stage image corresponding to the current land typeiminThe minimum value of the gray scale of each pixel in the processed early-stage image corresponding to the current land type is obtained.
4. The multi-temporal forest remote sensing image change monitoring method according to claim 1, wherein in step S4), a function expression for performing difference calculation on corresponding pixels in the normalized front and rear images one by one is as follows:
Figure FDA0002667001720000031
in the above formula, X'iIs the gray value, Y, of the pixel i in the normalized later-stage imagei' is the gray value of the pixel i in the normalized previous image, and e is the preset difference threshold.
5. The multi-temporal forest remote sensing image change monitoring method according to claim 1, wherein the specific step of mode filtering in step S5) comprises: and aiming at the image change image B or B', selecting one pixel in the image change image as a current pixel, then setting a rectangular window with the current pixel as a center and taking one pixel as a distance, and if more than five eighths of adjacent pixels in other pixels around the current pixel in the rectangular window have the same value, replacing the value of the current pixel with the value of the adjacent pixels.
6. The multi-temporal forestry remote sensing image change monitoring method according to claim 1, wherein the specific step of focal point statistics in step S6) includes: and aiming at the repaired image change image B1 or B1', selecting one pixel in the repaired image change image as a current pixel, then setting a rectangular window with two pixels as a distance by taking the current pixel as a center, counting the average value of the values of all the pixels in the rectangular window, and taking the average value of the values of all the pixels as the value of the current pixel.
7. The multi-temporal forest remote sensing image change monitoring method according to claim 1, wherein step S7) specifically includes the steps of:
s71) binarizing the image change maps B and B 'respectively to obtain binary maps C0 and C0' of preliminary change monitoring results corresponding to covered forest lands and uncovered forest lands;
s72) vectorizing the binary images C0 and C0 'respectively to obtain vector images D0 and D0';
s73) merging the vectorized vector images D0 and D0' to obtain a vector image D for preliminary change detection.
8. The multi-temporal forest remote sensing image change monitoring method according to claim 1, wherein the step S9) specifically comprises the following steps:
S91) inputting the vector diagram D of the preliminary change detection after the broken plaque is removed into a deep neural network model;
s92) taking the vector diagram D of the preliminary change detection as a further judgment object range, classifying and screening through a neural network model, and then removing unchanged image spots caused by image radiation errors in the diagram to obtain the vector diagram D1 of the final change detection.
9. The multi-temporal forest remote sensing image change monitoring method according to claim 1, wherein the step S10) specifically comprises the following steps:
s101) converting all surface data of the pattern spots in the vector diagram D1 of the final change detection into point data;
s102) extracting the gray value of the corresponding pixel according to the point data;
s103) abnormal value detection is carried out on the gray values of all the extracted pixels through boxline graph analysis, K-means clustering is carried out after the abnormal values are removed, the gray value of the minimum clustering center is used as the gray value after the clustering analysis, and the change reason is analyzed according to the size relation between the gray value after the clustering analysis and the gray value of the forest lands planted and uncovered.
10. A computer-readable storage medium, wherein the computer-readable storage medium stores a computer program programmed or configured to implement the multi-temporal forest remote sensing image change monitoring method according to any one of claims 1 to 9.
CN202010921885.4A 2020-09-04 2020-09-04 Multi-temporal forest remote sensing image change monitoring method Active CN112101159B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010921885.4A CN112101159B (en) 2020-09-04 2020-09-04 Multi-temporal forest remote sensing image change monitoring method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010921885.4A CN112101159B (en) 2020-09-04 2020-09-04 Multi-temporal forest remote sensing image change monitoring method

Publications (2)

Publication Number Publication Date
CN112101159A CN112101159A (en) 2020-12-18
CN112101159B true CN112101159B (en) 2022-05-24

Family

ID=73757343

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010921885.4A Active CN112101159B (en) 2020-09-04 2020-09-04 Multi-temporal forest remote sensing image change monitoring method

Country Status (1)

Country Link
CN (1) CN112101159B (en)

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112927252B (en) * 2021-04-12 2023-09-22 二十一世纪空间技术应用股份有限公司 Newly-added construction land monitoring method and device
CN113379620B (en) * 2021-05-18 2023-10-27 中国资源卫星应用中心 Optical remote sensing satellite image cloud detection method
CN113095303B (en) * 2021-06-04 2021-09-28 成都数之联科技有限公司 Model training method, forest land change detection system, forest land change detection device and forest land change detection medium
CN113421273B (en) * 2021-06-30 2022-02-25 中国气象科学研究院 Remote sensing extraction method and device for forest and grass collocation information
CN114119575B (en) * 2021-11-30 2022-07-19 二十一世纪空间技术应用股份有限公司 Spatial information change detection method and system
CN114821056A (en) * 2022-04-24 2022-07-29 国家林业和草原局华东调查规划院 AI technology-based forest and grass resource change automatic interpretation method in remote sensing image
CN114817616B (en) 2022-06-29 2022-09-16 四川省林业和草原调查规划院(四川省林业和草原生态环境监测中心) Forest accumulation amount continuous monitoring method, system and execution method thereof
CN115830448B (en) * 2022-11-30 2024-02-09 广州市地质调查院(广州市地质环境监测中心) Remote sensing image comparison analysis method based on multi-view fusion
CN116267382A (en) * 2023-02-16 2023-06-23 西南大学 Ecological composite utilization method for under-forest water regulation, soil conservation and carbon fixation
CN117315491B (en) * 2023-11-28 2024-02-20 泰安市绿威园林有限公司 Ecological red line early warning method for forestry engineering construction

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101976437B (en) * 2010-09-29 2012-10-03 中国资源卫星应用中心 High-resolution remote sensing image variation detection method based on self-adaptive threshold division
CN108846832B (en) * 2018-05-30 2021-06-15 理大产学研基地(深圳)有限公司 Multi-temporal remote sensing image and GIS data based change detection method and system
CN109657616A (en) * 2018-12-19 2019-04-19 四川立维空间信息技术有限公司 A kind of remote sensing image land cover pattern automatic classification method
CN110879969A (en) * 2019-10-21 2020-03-13 北京中科锐景科技有限公司 Forest land change detection method based on remote sensing image

Also Published As

Publication number Publication date
CN112101159A (en) 2020-12-18

Similar Documents

Publication Publication Date Title
CN112101159B (en) Multi-temporal forest remote sensing image change monitoring method
CN111709379B (en) Remote sensing image-based hilly area citrus planting land plot monitoring method and system
CN108846832B (en) Multi-temporal remote sensing image and GIS data based change detection method and system
CN106951836B (en) crop coverage extraction method based on prior threshold optimization convolutional neural network
US8615133B2 (en) Process for enhancing images based on user input
CN112164062A (en) Wasteland information extraction method and device based on remote sensing time sequence analysis
CN111191628B (en) Remote sensing image earthquake damage building identification method based on decision tree and feature optimization
CN104598908A (en) Method for recognizing diseases of crop leaves
CN111398176B (en) Water body water color abnormity remote sensing identification method and device based on pixel scale characteristics
CN108764284B (en) Classification and denoising method and system for high-resolution image of dead pine
CN108710864B (en) Winter wheat remote sensing extraction method based on multi-dimensional identification and image noise reduction processing
CN114581764B (en) Underground structure crack disease discriminating method based on deep learning algorithm
CN111476197A (en) Oil palm identification and area extraction method and system based on multi-source satellite remote sensing image
CN116559111A (en) Sorghum variety identification method based on hyperspectral imaging technology
CN115272848A (en) Intelligent change detection method for buildings in multi-cloud and multi-fog farmland protection area
Dang-Ngoc et al. Citrus leaf disease detection and classification using hierarchical support vector machine
CN111046838A (en) Method and device for identifying wetland remote sensing information
CN113033386B (en) High-resolution remote sensing image-based transmission line channel hidden danger identification method and system
CN116310913B (en) Natural resource investigation monitoring method and device based on unmanned aerial vehicle measurement technology
Wu et al. A Dense Litchi Target Recognition Algorithm for Large Scenes
CN107657246A (en) A kind of remote sensing image building detection method based on multi-scale filtering building index
CN115512159A (en) Object-oriented high-resolution remote sensing image earth surface coverage classification method and system
CN116310864A (en) Automatic identification method, system, electronic equipment and medium for crop lodging
CN115115948A (en) Forest land information fine extraction method based on random forest and auxiliary factors
CN114596489A (en) High-precision multisource remote sensing city built-up area extraction method for human habitation index

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