JP6273473B2 - Forest information management device - Google Patents

Description

  The present invention relates to an apparatus for recognizing and managing forest trees. In particular, with regard to devices that use multispectral images taken from the sky on artificial satellites or aircraft, etc., by recognizing and managing the individual positions of trees using latitude and longitude information, integration of different information sources and reliability The present invention relates to a device for maintaining and managing high forest information.

Conventionally, as forest information, there are the basic forest map, forest plan map, forest book, forest group map (operation implementation plan map), and small group map outlined in (1) to (5) below, which have been widely used. These were the standard information for forest GIS.
(1) Basic map of forest A map showing the topography and the boundary between forests (forest group boundary). This is a topographical map that is used as basic data for forest maintenance and that also fills in the land use status such as contour lines, mountain villages and roads.
(2) Forest plan map In addition to the private forest area, the forest boundary indicating the forest book area is described, and it becomes a document for identifying personal property. In addition, when logging within the forest plan map area, “Notification of logging and logging after felling (submitted to each municipal office)” based on the Forest Law, and “Land Development License ( Forest Management Division Jurisdiction) ”is necessary.

(3) This is a ledger that records the owners, tree species, forest age, area, etc., divided into forest groups that planted forestry mountains, but is a document created for the purpose of supporting the establishment of forest plans. It is often created by indirect surveys using photographs and interviews, and no actual measurements or confirmations have been made.
(4) Forest group diagram Considering the location of the forest and the convenience of operations, it is a drawing that describes the unit of forest section set for each municipality, with character boundaries and ridgelines as boundaries, and the start and end points of life trails and forest roads Have been described.
(5) Small group map The forest group is subdivided and classified according to tree species and forest age (tree age). Forest information is described in detail in the above (1) to (5), and FIG. 2 illustrates a forest subgroup diagram and a forest subgroup branch number diagram that is the minimum section.
What should be noted here is that the trees are not individually recognized and managed even in the forest sub-branch branch map, which is the smallest category, and even in the GIS that makes these (1) to (5) a database, individual trees are recognized. There is no management.

Prior art related to forest information management is described in patent technical literature and non-patent technical literature.
Patent document 1 is called “forest information system”, and the method is a forest information processing system that obtains the distribution of tree species in a forest area using a color image obtained by photographing the forest area from the sky, and converts the color image into a black and white image. The crown shape image creating means for creating the crown shape image by obtaining the crown shape from the light / dark inclination information, and obtaining the wavelength spectrum of the color image portion corresponding to the crown portion of the crown shape image, for each tree species obtained in advance This is a forest information processing system having tree species specifying means for specifying the tree species of the crown portion with reference to the wavelength spectrum of the tree.
The characteristics of this technology are as follows:

(1) According to the rise of the crown, the luminance is highest at the apex portion of the crown and the luminance change is smallest at the apex portion. The so-called watershed algorithm uses the change in brightness due to the roundness of the crown, where the brightness decreases toward the periphery of the canopy and becomes the lowest at the border between adjacent crowns. The marker is set at the center of the part where the change in brightness (gradient) is the smallest, and the initial marker at the top of each crown is grown in four directions along the change in brightness (gradient). The point where the marker of the crown to collide with the crown is determined as the boundary of the crown. The crown shape is extracted by growing a region along the luminance change with the crown top having a small luminance change (gradient) as a starting point.

(2) Subsequently, the tree shape is obtained by superimposing the canopy shape image and the original color image input to the computer, and performing spectrum analysis.
It is limited to using gradient, which is a kind of differential operation for light and dark, and because the tree crown is extracted by the single peak of light and dark, it is weak against noise components and the crown is closed Applications are limited to no coniferous forests. Furthermore, there is no particular novel classification of tree species, but because it is used for classification of extracted crowns, it is not possible to extract tree boundaries using differences in tree species, and applicability to broadleaf and chaotic forests. There is a limit.
The biggest difference from this patent is that individual trees are not identified and managed by absolute position information.

Patent Document 2 is referred to as a “tree number calculation method and tree number calculation device”. This method calculates the number of trees included in a forest from the image data obtained by photographing the forest from above and the geographical information of the forest for each tree type. This is a calculation method that extracts each tree vertex and each tree crown from the image data, classifies the tree species from the spectral reflection characteristics of the sunny part of the tree crown, obtains tree species classification image data for each tree vertex, and further includes the trees included in the area This is a method for calculating the number of trees by tree type. According to the specification, the extraction accuracy of the crown from which the number of trees starts is as follows: `` The extraction of the crown is performed using the fact that the edge of the crown (called the crown edge) is darker than the crown part. Can be achieved by a well-known blur following algorithm (Va11ey fo11owing a1gorithm) and automatic mask processing, and by extracting each canopy, the pixels corresponding to each extracted canopy are accumulated, The area of the canopy can be determined, "and depends on the known Va11ey fo11owing a1gorithm.
Furthermore, Patent Document 2 does not identify and manage individual trees based on absolute position information.

Patent Document 1 uses Watershed Algorithm and Patent Document 2 uses Va11ey fo11owing a1gorithm for tree crown extraction, but Non-Patent Document 3 has verified the accuracy of these two algorithms. Fig. 4 shows a copy of the field verification results. The detection error of the crown is + 10% to + 15% for watershed algorithm and Va11ey for conifers.
fo11owing a1gorithm has an error of -5% to -20%, Watershed Algorithm is -18% for broadleaf trees, Va11ey
fo11owing a1gorithm has an error of + 5%. Fig. 3 shows an example of a crown in a Japanese broadleaf forest. (1) The crown 1122 in the crown recognition 1 is correct, but (2) the crowns 2-1 to 3 123-5 in the crown recognition 2 are algorithmically used. Or (3) It recognizes to multiple like the crown recognition 3.

  Patent Document 4 refers to an “image cluster analysis device”, which relates to a device for recognizing and selecting forest tree species from a multispectral image of the ground surface photographed from above, introducing unsupervised cluster analysis technology to improve the discrimination ability, This is a technique for determining a tree type for multispectral data for which the corresponding tree type is unknown using a ground measurement device, and latitude and longitude are used as the absolute position of the individual tree for the association. Patent Document 4 is a patent by the same inventor as that of the present invention, and identifies tree species with absolute position information with high accuracy. The present invention identifies and manages individual trees based on absolute position information as an adjacent technique of Patent Document 4, and the description of the multispectral image region classification processing 138 follows Patent Document 4.

The method of Non-Patent Document 5 is a research paper that identifies 10 types of trees using 8-band multispectral satellite images. The reference spectrum is measured for each tree type, and the forest spectrum and It is a technology that identifies forests by tree type by comparison. Classified as supervised cluster analysis. However, identification management of individual trees based on absolute position information is not performed.

Japanese Patent No. 4441883 Patent No.4858793 Comparison of Indivisual Tree Crown Detection and Deliniation Method, Yinghai Ke, Lindi J. Quackenbush, Environmental Resources and Forest Engineering, State University of New York College of Environmental Science and Forestry, ASPRS 2008 Annual Conference Portland, Oregon April 28-May 2,2008 Japanese Patent Application No. 2012-239685 Commercial Timber Tree SpeciesIdentification Using Multispectral WorldView-2 Data, HamdanOmar, DigitalGlobe 8-Band Research Challenge

The conventional forest information management has the following first to fifth defects, and does not address forest management issues of about 500 ha or less. That is,
First, forestry management is required to cope with long-cutting and multi-phase forestation, but individual information of trees is necessary, and conventional basic information (basic forest map, forest plan map, forest book) , Forest group diagram, small group diagram) is based on the premise of clear-cutting for each area and cannot be handled.
Second, roadside forestry with forestry machines has been introduced in recent years, but individual information in units of trees that is finer than conventional basic information is required, and conventional basic information cannot be used.

Thirdly, conventional forest information management is a graphic information that represents the area of basic information (basic forest map, forest plan map, forest book, forest group diagram, sub-group diagram), so graphic information must be systematized at the core. Since GIS is not obtained, GIS is used. GIS is an advanced data base that integrates graphics and relational databases, so information maintenance is not easy.
Fourthly, management interests for forestry managers are forest asset conservation management as assets, and individual information of trees is necessary to accurately grasp assets. Conventional basic information is too sparse, even a relatively large forestry manager of 500ha can fit all the area of 45cm x 45cm in 1/5000 drawings and cannot manage 200,000 to 500,000 trees . In addition, the conventional tree individual recognition technique has a large error, and cannot be used for recognition management based on the position reference as it is.

Fifth, there is forestry management with double-phase forests, and Japanese forests are originally diverse in tree species. If you try to use this to manage profitable mountain products and commercialize them, conventional basic information does not contain information. Insufficient to handle.
All the above problems are caused by the fact that trees cannot be recognized individually and managed with high accuracy.

In the present invention, the above-mentioned problem has been fundamentally solved by performing forest information management in units of individual trees. The method of managing trees individually is the optimal solution because the method based on latitude and longitude is a general-purpose standard and the information of measurement systems such as GPS can be used directly. However, all of the following issues must be solved in order to manage trees individually by latitude and longitude.
(1) A high-resolution image with high positional accuracy (latitude and longitude) and no distortion can be obtained.
(2) A means for accurately recognizing individual trees is established. There is a certain level of error in tree position measurement from images only, but it is possible to obtain a mobile measurement means with high position accuracy on the spot for practical purposes of an algorithm that reduces this error and for the purpose of eliminating the error. about.
(3) An algorithm that can accurately calculate the absolute position of a tree from an image with a practical amount of computation exists.
(4) There are a high-resolution image multispectral image and an algorithm that can specify a tree type corresponding to the absolute position of the tree from the image with a practical amount of computation.
The positional accuracy referred to in (1) to (4) is for the purpose of specifying trees individually, and is therefore at least half of the tree interval. Usually, 1600 trees are planted per ha, and thinning is carried out to make 400 trees per ha, resulting in 6.25 m2 / tree, and the minimum required position accuracy is 1.8 m.

In the present invention, (1) to (4) are all realized and solved by the new means.
First, a high-resolution multiband multispectral image was used as the image. The satellite image is used because the characteristics such as parallax of the entire image are uniform. In particular, the latitude and longitude accuracy is high, and the image size is large, so that there are advantages that the accuracy of the ortho image can be obtained in a wide range. With commercial satellite images, panchromatic 50 cm and multispectrum 1.6 m can be obtained, so the minimum required position accuracy can be satisfied over the entire screen area. In addition, the fact that the image characteristics are uniform throughout the image and the number of spectral bands is large makes it possible to improve and optimize the algorithms (4) and (5). Here, a high-resolution multiband multispectral image is used for ease of acquisition, but an airborne sensor may be used.

Secondly, it has been found that even when image information with high positional accuracy is obtained in (1), the accuracy of the conventional tree crown information obtained only from the image is insufficient. In addition to improving accuracy with the new method, a means to obtain supplementary information and correction information on site will be prepared. In this case, it is necessary to efficiently collect information on site with the same positional accuracy as (1). For this purpose, high-speed measurement with an accuracy of 1 m or less is possible using DGPS, and the mobile terminal can be operated in conjunction with the processes (1), (3), and (4).
Third, in order to calculate the absolute position of the tree from the image, in addition to the first means and the second means, a local area that maximizes the amount of information (entropy) with respect to the panchromatic information (panchromatic image) of the image The entire image is scanned with a standard histogram leveling filter, the crown is emphasized, the crown is detected by correlating with the crown shading model that reflects the shooting direction and solar radiation direction, and the tree position is determined from the pixel coordinates in the image. Expressed in latitude and longitude. This method uses image information as effectively as possible for the purpose of canopy recognition. Further, since it is an integral system operation rather than a conventional differential system operation, it has a strong characteristic against noise and improves discrimination ability.

Fourth, when it becomes possible to identify and manage forest trees individually using position information based on latitude and longitude, it is essentially different from area management such as a forest group chart, and “the tree does not walk, it does not change suddenly” It is possible to store information for individual trees by utilizing the above, and the following advantages (1), (2), and (3) occur.
(1) It is generally difficult to obtain the position of a tree with a 100% accuracy from a temporal information accumulation crown based on latitude and longitude, and the difficulty is particularly high in a broadleaf tree. However, immediately after thinning, the canopy opens, making it easier to identify, and broad-leaved trees have seasonal changes, so if you manage the trees individually in absolute positions, you can accumulate information over time, and the information quality improves.

(2) On-site information storage based on latitude and longitude criteria As a macro information source of forests, information is extracted from high-resolution multispectral images by image processing technology, and micro-site information obtained by DGPS is managed and integrated by latitude and longitude. I can do it.
(3) Grasping of statistical secular change by individual recognition When measuring the growth amount of trees, it is impossible to measure the total number of trees, so it is necessary to make statistical estimations. Need to be the same every year. When performing statistical processing, the accuracy of the statistic is improved because the sample can be specified by the absolute position.

Fifth, applying the method of Patent Document 4 Japanese Patent Application No. 2012-239685, using a high-resolution multispectral image of the forest, the tree species is specified with latitude and longitude coordinates with high accuracy, and the individual tree is combined with the canopy information. Added latitude and longitude information. That is, in order to solve the following defects (1) to (4) of the conventional “supervised cluster analysis”, a new method “unsupervised cluster analysis” is introduced.
(1) In supervised cluster analysis, it is necessary to predetermine the spectrum of an object such as a tree species as a reference for comparison. When collecting a sample of an object and measuring the spectral distribution in advance to obtain a reference spectral distribution, the reference spectral distribution is measured in a different environment from the actual image. Occurs and the identification ability decreases. In addition, when selecting multiple image areas of an object from a real image and performing statistical processing to obtain a reference spectrum distribution, the spectrum that is not the original object in the process of collecting the reference spectrum from the object image Incorporating the distribution is unavoidable and the discrimination ability is reduced.

(2) In a forest where a plurality of vegetation is mixed, the spectral distribution is mixed and the discrimination ability is lowered. In particular, when bamboo forests invade forests, it is difficult to detect bamboo forests because they invade in a mixed manner with existing tree species.
(3) In forests where the tree species that exist are not limited, there are those that do not have a tree species reference spectrum, so it is not possible to grasp all the forest resources in the forest. It cannot be used for diversified uses such as collecting rare plants, wild plants in chaos, and collecting mushrooms.
(4) In the high-resolution satellite image, the pixel size on the ground becomes small, and the situation in which the object is not uniform occurs. Therefore, a noise component is generated by mixing a component that is not the original reference spectrum, and the identification capability is reduced. The advantages of high-resolution satellite images cannot be fully utilized.
For the above (1) to (4) defects or problems, Patent Document 4 Japanese Patent Application No. 2012-239685 takes the following solutions (1) to (3), which are also applied to the embodiments of the present invention. .
(1) Those having similar pixel spectral characteristics are grouped together from the target screen, and the reference spectrum is autonomously acquired and classified without being given from the outside. As a result, all tree species in the forest can be classified, the spatial resolution and spectral resolution of the high-resolution satellite can be utilized to the maximum, and the ground surface can be classified in a state where spectral data is not mixed.
(2) For all the pixels of the target image, the similarity with all the other pixels is obtained by the inner product operation between the pixel vectors. Solved.
(3) Since grouping is based on mutual similarity between pixels of the target image, it is necessary to determine the tree species corresponding to each group, but the actual tree species is made to correspond to the pixel group for which there is no corresponding tree species information. A device for identifying portable attribute data was devised.

  As described above, according to the present invention, forest trees can be managed individually, so that not only can cope with long cutting periods, multi-phase forests, roadside forestry, but also the forest information structure is managed centrally by latitude and longitude. This simplifies and makes maintainability easier. In addition, as there is the word “not seeing the forest by looking at the trees”, the feature of forests is that there is a recognition means gap between macro (forest) information and micro (standing trees) information. Information gaps lead to wasteful and non-optimal actions, further hindering efficient accumulation of information and reducing labor productivity. The latest information technology has been able to actively close this gap.

Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the drawings. The present invention is not limited to the following description, and various modifications can be made without departing from the scope of the invention.
FIG. 1 is a diagram showing an overall configuration of a forest information management apparatus according to the present invention, which includes a forest information management apparatus fixing unit 100 and a portable attribute data specifying apparatus 110. The forest information management device fixing unit 100 may be a commercially available high-performance personal computer at the time of the filing of this patent application. The arithmetic device A 101, the multispectral image data device 102, the attribute group data device 103, the forest information DB device 104, and the operation information data apparatus
109, DEM data device 126, forest management information device 127, and data interface device A 105, and operator 106 operates through a monitor, keyboard, and mouse. The multispectral image data device 102, the attribute group data device 103, the forest information DB device 104, the operation information data device 109, the DEM data device 126, and the forest management information device 127 can be configured by a storage device such as a normal HDD.

A multispectral image is composed of pixels corresponding to vertical and horizontal two-dimensional coordinates, and each pixel is a multidimensional vector array corresponding to the number of multispectrals. The pixel group to which the pixel belongs corresponding to each pixel is assigned in the multi-spectral image pixel / pixel group correspondence table 233 in FIG. The multispectral image and multispectral image pixel / pixel group correspondence table 233 is stored in the multispectral image data device 102.
An attribute code that defines the attribute is assigned to each pixel group of the multispectral image by the pixel group / attribute correspondence table 234. In the case of forest, the attribute code is a tree species code. There are a plurality of pixel groups, and one attribute code is assigned to each pixel group. The same attribute code may be assigned to a plurality of pixel groups.
The contents of the multispectral image data device 102, the attribute group data device 103, and the forest information DB device 104 are exported to and imported from the portable attribute data specifying device 110 via the data interface device A 105. At the start of the calculation, data allocation of the multispectral image pixel / pixel group correspondence table 233 and the pixel group / attribute correspondence table 234 is not performed. Operation information data device
109 stores operation information 120 and disaster information 121, and stores information obtained by artificially changing the position information of trees including thinning, logging, and afforestation, and information in which the position of the trees has been changed due to landslides.

Data received by the portable attribute data identification device 110 from the data interface device A 105 of the forest information management device fixing unit 100 via the data interface device B 115 is the multispectral image data device 112, the attribute group data device 113, and the forest information Stored in DB device 114. The data interface device A 105 and the data interface device B 115 can be configured by a USB memory. The DGPS 116 is a differential GPS device for precise positioning, and can be positioned with a position accuracy of 1 m or less and is publicly implemented. The components of the portable attribute data identification device 110 other than the DGPS 116 and the laser range finder 118 are publicly implemented as a tablet PC. The portable attribute data identification device 110 uses a function for assigning attribute data to a pixel group of a multispectral image whose attribute data is undetermined by actual measurement in the field, and measured values of the DGPS 116, the direction sensor 119, and the laser rangefinder 118. There is a function to measure the position of the irradiation destination including the tree of the laser rangefinder 118 in latitude and longitude.

FIG. 5 shows the overall information flow of the forest information management apparatus according to the present invention, and FIG. 6 shows the overall configuration of the processing. The following description will be given in a manner that will explain both figures hierarchically in more detail. In FIG. 5, a new image 107 including a multi-band multispectral high resolution satellite image is read into the forest information management device fixing unit 100 and is orthorectified by a publicly implemented technique using the reference GCP 108 at processing block 001 130. Among the orthorectified images, the multispectral image is obtained by the cluster analysis (FIGS. 5 and 6) of the processing block 010 140 of the multispectral image region classification process 138 and the tree species corresponding to the latitude and longitude. For those whose tree species are unknown, multi-spectral image pixel / pixel group correspondence table 233, pixel group / attribute correspondence table 234, tree information DB
136,
Multi-spectral image portable attribute data identification device
110 is transferred via the data interface device A 105 to determine the unknown tree species. The contents of the tree information DB 136 include the items shown in FIG. Each tree is identified by latitude and longitude, and corresponding information is stored. Furthermore, it is possible to link to the character information file 166, the image information file 167, the audio information file 168, and the moving image audio information file 169 in units of trees via the file link information.

Among the images orthorectified in the processing block 001 130, the panchromatic image 139 is subjected to histogram leveling processing in the processing block 0111 141, and the light / dark level adjustment is performed so that the tree crown can be easily extracted. In processing block 012 142, tree shading model generation processing is performed in processing block 012 142 according to the imaging condition 135 associated with the new image 107. Tree Shading Model models a tree with a spheroid and calculates the apparent shadow pattern of the tree based on the solar radiation angle, viewing angle, and spheroid parameters. A tree position is calculated in a processing block 013 143 from the tree Shading Model generated in the processing block 012 142 for the panchromatic image 139 whose histogram is leveled in the processing block 011 141, and a tree position is obtained by latitude and longitude. Since the tree position calculated here may have an error due to an error in crown recognition, the portable attribute data specifying device
110 gives a correction function.
In FIG. 5, a portable attribute data identification device
All parts other than 110 are processed by the forest information management device fixing unit 100.

FIG. 8 is a diagram showing the principle of processing block 010 140 cluster analysis processing. (1) Natural color cluster analysis is to identify attribute A and attribute B by a pixel vector based on RGB three primary colors, and (2) 8-band multi-spectral cluster analysis is identified by an 8-band pixel vector instead of RGB three primary colors. . As a method, when the inner product of the pixel vector of the attribute A and the pixel vector of the attribute B normalized with the norm and the pixel vector of the recognition target is obtained and the deviation from 1.0 is within the threshold, the recognition target pixel vector is the attribute A or Identify as attribute B. A different number of spectral bands can be used in place of the 8-band spectral data. FIG. 8 (3) is a conceptual diagram of 12-band multispectral cluster analysis. Generally, if linear independence is strong between vectors constituting a pixel vector, the amount of information increases and the discrimination ability increases as the vector dimension increases.

Linear independence is high between each element of a pixel vector by a multispectral band with little overlap of spectral bands. (To be precise, it is necessary to determine the cross-correlation coefficient, and it depends on the reference substance. However, the linear independence is the highest when there is no overlap of the spectrum band.) Between the elements of the pixel vector with high linear independence The vector element obtained by the non-linear operation can have linear independence with respect to the original pixel vector. The following indicators (1) to (9) can be added to the original measured pixel vector to have linear independence. Therefore, adding to the base pixel vector to increase the dimension improves the discrimination ability. It can be made. In the following formula, NIR: near infrared, R: red, MIR: mid-infrared, SWIR: short wavelength infrared, for example, a = 0.96916, b = 0.084726, L = 0.5. It is.

(1) Ratio Vegetation Index (RVI) That is, RVI = NIR / R
(2) Normalized Difference Vegetation Index (NDVI) That is, NDVI = (NIR-R) / (NIR + R)
(3) Normalized Difference Wetness Index (NDWI) That is, NDWI = (SWIR-MIR) / (SWIR + MIR)
(4) Green Vegetation Index (GVI) That is, GVI = (NIR + SWIR) / (R + MIR)
(5) Perpendicular VI (PVI) That is, PVI = (NIR-aR-b) / (1 + a ^ 2) ^ (1/2)
(6) Weighted difference VI (WDVI)
That is, WDVI = NIR-NIR (soil) / R (soil) * R,
NIR (soil) = aR (soil) + b
(7) Soil Adjusted VI (SAVI) That is, SAVI = (1 + L) (NIR-R) / (NIR + R + L)
(8) Soil Adjusted Ratio VI (SARVI) That is, SARVI = NIR / (R + b / a)
(9) Transformed Soil Adjusted VI (TSAVI)
TSAVI = a * (NIR-a * Rb) / (R + a * NIR-a * b)

Processing block 010 140 In the cluster analysis process, correlation operations between pixels of the target image are performed to group together those having high correlation, but this calculation requires enormous operations. 10,000 pixels on a side of 10,000 pixels X
10,000 for a 12-dimensional vector pixel vector of 10,000 pixels
X 10,000 X 10,000 X 10,000 / 2 sets of correlation calculations are required, the pixel vector is 12-dimensional, 100FLOPS is required for one set of pixel correlation calculations, and the calculation device is 100GFLOPS. It takes 58 days and is not realistic.

A method for solving this problem is shown in the correlation calculation area setting method of the processing block 010 140 cluster analysis processing of FIG. Hereinafter, a case where the target is a forest will be described as an example, but the same applies to a target other than a forest. First, Normalized forest area 151 from multispectral image 150
Difference Vegetation Index (NDVI), Normalized
Extracted by spectral calculation including Difference Wetness Index (NDWI). Next, from the forest region 151, regions for which correlation calculation between pixels is performed are selected as a target region 1 152 to a target region i 157 to a target region n 158. The purpose is to collect pixel vectors corresponding to the attribute data of the forest region 151 while reducing the amount of correlation calculation between the pixels. Therefore, the target region is set to a portion that is visually determined to represent a plurality of target regions and the attributes of the forest region 151 (tree species in the case of forest).

The target region i 157 in the forest region 151 will be described below as an example in “Method of calculating inter-pixel correlation in multispectral image region classification processing 138” in FIG. The correlation calculation between the pixel vectors is performed in the range of the target area i. After obtaining the attribute data for the target area i, the attribute data for the other target areas is obtained in the same manner, and then all the target areas are obtained. The method of integrating the attribute data of If one side of the target area i is set to 100 × 100 pixels, for example, the number of correlation operations between pixels is reduced to 500,000 sets. In FIG. 10A, correlation calculation is performed between all the pixels constituting the target area i, and a combination of pixels having the highest correlation is selected and extracted as a combination of Rank i = 0.

Next, in FIG. 10 (2), the correlation calculation is performed on all the other pixels constituting the target area i for the combination pixel with Rank i = 0, and the correlation is higher than a certain threshold (for example, the inner product is 0.9995 or more) Are connected and grouped as Rank i = 1 pixels with respect to Rank i = 0 pixels. Further, in FIG. 10 (3), the Rank i = 1 combined pixel is subjected to correlation calculation with all other pixels constituting the target area i, and the correlation is higher than a certain threshold (for example, the inner product of 0.9995 or more). Connect and group as Rank i = 2 pixels for Rank i = 1 pixels. In FIG. 10 (3), Rank i = 3 is similarly grouped. This operation is repeated in advance (for example, 4 to 5 times), and the grouping started from the pixel selected in FIG.

The operation described in the previous section is applied to the remaining pixels of the target region i that are not grouped in the previous section, and a pixel group different from the previous section is generated. This operation is repeated to generate M groups from the pixels in the target area i. FIG. 11 is a diagram illustrating a concept of processing for integrating pixel groups for each target region in a multispectral image. A pixel group is generated in each target area from the target area 1 to the target area M in FIG. The method shown in FIG. 10 does not guarantee the independence of the pixel groups in each target region and the pixel groups between the target regions, and pixels having the same attribute are grouped into different groups. The similarity between pixel groups can be obtained by averaging pixel vectors representing each pixel group (for example, averaging and normalizing the pixels in the group; this is referred to as a representative vector). When it is equal to or larger than the inner product (0.9995), it is determined that it corresponds to the same attribute, integration between groups is performed, and an integrated pixel group j for the integration target region shown in FIG. 11 can be obtained. A pixel vector representing each integrated pixel group (representative vector) corresponds to each attribute serving as a reference for region classification of the multispectral image.

The overall structure of the multispectral image region classification process 138 is shown in FIG. The processing of the processing block 020 160 is performed in the forest information management apparatus fixing unit 100. This is pre-processing in the forest information management apparatus fixing unit, and is processing for generating representative pixel vectors for each integrated pixel group described up to FIG. 11 for the input multispectral image. At this stage, pixel vectors of most pixels of the input multispectral image are classified into integrated pixel groups, but it is not determined which attribute such as a tree species corresponds to each integrated pixel group. Information that is known in advance (for example, a priori information that a tree species called XX grows in this area) and has a corresponding attribute relationship such as each integrated pixel group and tree species is multi- The region classification of the spectrum image is performed (this function is also provided in the portable attribute data specifying device 110 and will be described later). Processing block 021 161 defines the integrated pixel group and tree species that do not have a corresponding attribute, such as a portable attribute data identification device.
Using 110, the integrated pixel group and the attribute correspondence such as tree species are determined by visual confirmation on site. In the processing block 022 162, the forest information management apparatus fixing unit 100 completes the multispectral image region classification processing 138 by reflecting the attribute correspondence such as the integrated pixel group and the tree species associated in the processing block 021 161.

Hereinafter, the processing block 020 160, the processing block 021 161, the processing block 022 162, and the processing block 023 163 of FIG. 12 will be described in detail with reference to FIG.
FIG. 13 shows the contents of the processing block 020 in FIG. In processing block 030 170, the target area for performing cluster analysis from the target image is limited for the purpose of reducing the amount of calculation. For forest tree species classification, only new areas are extracted. Details are shown in FIG. Processing block 040 180 performs land water area exclusion processing. Publicized Normalized
The difference water index can be used to determine the threshold value. Processing block 041 181 excludes areas other than plant areas by excluding areas where the chlorophyll activity index NDVI is below the threshold. In processing block 042 182, the bare ground region is similarly excluded from the determination based on the magnitude of the threshold value. Publicly implemented Soil
Soil index including Adjusted Ratio VI (SARVI) can be used. Processing of the processing block 040 to processing block 042 180 to 182 is out of order, and other processing may be added as necessary.

In processing block 043 183, a plurality of target areas for performing correlation calculation between pixels are set from the area for performing cluster analysis. The processing described in the explanation of the figure, and after selecting a region (forest region 151 in the case of FIG. 9) where cluster analysis is performed from the multispectral image 150, a plurality of target regions 1 to target regions i to target regions n are further selected. It is a process to set. The area can be specified by a publicly practiced method such as inputting the coordinates on the diagonal line of each target area while operating the mouse on the monitor.
In processing block 031 171, a pixel vector of a target region for cluster analysis is generated and normalized. In addition to the measured spectrum band by the multispectral sensor (hereinafter referred to as the measured pixel vector), the pixel vector can be added as a vector element with the result of nonlinear calculation between the measured spectra as an additional band, if necessary. A new pixel vector is redefined. (Hereinafter, the pixel vector is referred to as an extended pixel vector or simply a pixel vector.) Since the light and dark information is harmful for the purpose of spectrum analysis, the pixel vector is normalized so that only the spectral distribution information is obtained.

At processing block 032 172, the target area for cluster analysis in the target image is classified into pixel groups based on the similarity of pixel vectors. The contents of the processing block 032 172 will be further described with reference to the drawings. A processing block 050 190 indicates that the representative areas divided into N pieces in FIG. 9 are sequentially processed. Processing block 051 191 is processing for enumerating and extracting combinations determined to be similar for all pixel vectors constituting the representative region for the selected representative region, and the contents will be described in detail with reference to FIG.

Processing block 060 200 calculates all combinations of inner products of pixel vectors among all the pixels that make up the selected representative region. In order to determine the similarity between pixels, the similarity between all pixel vectors is grasped as an inner product. The processing block 061 201 calculates the inner product distribution (histogram) of the pixel vectors among all the pixels constituting the selected representative region. Next, in processing block 062 202, a threshold value of an inner product for determining that the pixel vectors are similar from the above distribution (histogram) is obtained. As a threshold standard, one method calculates the standard deviation of the inner product from 1.0 (referred to as inner product deviation), and the range from 0.0 to the standard deviation contains the number of 3σ pixels centered on the standard deviation. As an alternative, 0.0 means an inner product deviation means that the pixel vectors match.Therefore, starting from the inner product deviation 0.0, the histogram is viewed in the increasing direction of the inner product deviation. There is also a method in which the inner product deviation that causes the inflection point of the histogram is used as a threshold value. The processing block 063 203 enumerates and extracts all the combinations determined to be similar with the threshold value of the processing block 062 202 for all the pixel vectors constituting the selected representative region.

Returning to FIG. 15, processing block 052 192 groups all pixel vector combinations determined to be similar for the selected representative region. Since the processing of the processing block 051 191 extracts only a combination of two similar pixels, these are connected to each other to be combined into a pixel group. Details are described in FIG. Although the concept has been described with reference to FIG. 10, a specific procedure will be described below.

Processing block 070 210 sequentially processes the M pixel pairs (combinations) of all pixel vectors determined to be similar (in processing block 063 203) for the selected representative region. In processing block 071 211, the first isolated pixel pair is extracted from all the pixel vectors determined to be similar. The isolated pixel pair here refers to a pixel that is similar only between two pixels (FIG. 10 (1)). Next, processing block 072
The first isolated pixel pair extracted in 212 is connected Rank i = 0. The connection Rank indicates the level of grouping of similar pixels, and what is related only between two pixels as shown in FIG. 10A is called Rank i = 0. On the other hand, pixels determined to be similar to the two pixels in FIG. 10A are grouped and connected as Rank i = 1 as shown in FIG. This processing is performed until there is no connection rank i + 1 pixel pair that can be connected to the connection Rank i pixel pair among all the subsequent unconnected pixel pairs selected in the processing block 075 215. (The connected pixel is no longer an isolated pixel pair) When there is no more connection Rank i + 1 pixel pair that can be connected, the processing Rank 076 216 adds +1 to the connection Rank i and connects the next Rank. As described above, the connection is repeatedly performed and grouped as shown in FIG. 10 (3) and FIG. 10 (4), but the similarity with the pixel of Rank i = 0 decreases as the number of connected Ranks increases. Therefore, in process block 076 216, an upper limit k is set for the number of connected ranks (empirically, 5 is appropriate for leaves), the process in process blocks 073 to 76 is terminated, and the process returns to process block 071 211 to return to the group. The processing of the processing blocks 071 to 76 is performed as a different pixel group generation operation on the remaining pixel pairs. This is repeated until there are no more groupable groups for all M pixel pairs.

Processing block 052 Since the processing of 192 connects similar pixel pairs one after another, there is no guarantee that the pixel vector of Rank i = 0 is the center of the same pixel group even if it belongs to a united pixel group. Yes. Thus, at processing block 053 193, pixel vectors representing all pixel vectors of the same pixel group are recalculated. The recalculation method includes a method of obtaining and normalizing an average value of all pixel vectors of the same pixel group, and a method of recalculating the representative pixel vector by excluding a pixel vector having a large deviation from the representative pixel vector from the group. is there. Hereinafter, the representative pixel vector recalculated in this way is referred to as a nucleus vector.

In the processing of processing blocks 050 to 53, the grouping of pixels for the N representative regions and the nucleus vector thereof have been obtained, but not only the same attribute of the multispectral image 130 corresponds to a plurality of pixel groups in different representative regions. Even in the same representative region, a plurality of pixel groups may correspond to the same attribute. The processing block 054 194 solves this problem, the details of which are described in FIG. First, in processing block 080 220, for each of the N representative regions, the groups are arranged in descending order of the number of pixels, and are arranged from the excellent pixel group. Next, at processing block 081 221, the boundaries of the representative areas are removed, the N representative areas are merged into one area, and the core vectors of the groups in each area are aligned in descending order of the number of pixels.

Processing of processing blocks 082 to 86 is processing block 081
This is a process for evaluating the similarity between the pixel groups summarized in 221 by the inner product between the nucleus vectors and integrating the pixel groups. At processing block 083 223, the correlation (inner product) between all the nuclear vectors is calculated. If the maximum correlation (inner product) value between all the nuclear vectors is smaller than the threshold value at processing block 084 224, the nuclear vectors are determined not to be similar to each other, and the pixel group integration ends. If any correlation (inner product) between the nuclear vectors is larger than the threshold, the image groups represented by the nuclear vectors are judged to be similar, and the image groups are integrated into the image group integrated in processing block 085 225. Recalculate the nuclear vector. Further processing block 086
At 226, the nuclear vectors are arranged in descending order of the number of pixels in the group. The processing of the processing blocks 085 to 086 is continued until the processing block 084 224 is established.

Hereinafter, the description will return to the processing block 033 173 in FIG. Since it is necessary to specifically assign attribute data (tree species in the case of forest) to all representative pixel vectors (nuclear vectors), processing is performed sequentially. In processing block 034 174, if the attribute of the processing target representative pixel vector (nucleus vector) is determined, the processing block 035 175 corresponds to the pixel group of the pixel group / attribute correspondence table 234 in FIG. Register attributes. Here, FIG. The table structure which prescribes | regulates an attribute (in the case of forest) in a multispectral image is shown. Multispectral image Pixel / pixel group correspondence table 233 describes the multispectral image 150 in two-dimensional coordinates (H, V) of the H coordinate axis 230 and the V coordinate axis 231 and groups each pixel as a pixel group. Is assigned a group number (Grp j, Grp j-1 in FIG. 19 (1)). The correspondence between the group number and the attribute is defined in the pixel group / attribute correspondence table 234, but in the processing block 034 174, the correspondence is not determined except for the correspondence between the pixel group and the attribute based on the prior information. .

In processing block 036 176, the multi-spectral image pixel / pixel group correspondence table 233 shown in FIG. 19 and the pixel group / attribute correspondence table 234 whose partial correspondence is undetermined are output as portable attribute data specifying device input data. To do. For output, the data interface device A 105 of the forest information management device fixing unit 100 can be used and can be output from a USB memory. A multispectral image may be output for reference display. The portable attribute data specifying device 110 reads this data via the data interface device B 115.

The portable attribute data specifying device 110 completes the pixel group / attribute correspondence table 234. FIG. 20 shows the operation and function thereof. The system configuration is the portable attribute data specifying device 110 of FIG. 1, and can be realized by the built-in functions of the tablet PC 241 as shown in FIG. 20 except for the DGPS 116 and the laser rangefinder 118. The current position 244 of the portable attribute data specifying device 110 is measured by using the DGPS 116 installed on the upper side of the support column 240 as the latitude and longitude. When DGPS is used, it can be measured with an accuracy of 60 cm. DGPS is publicly implemented as Hemisphere A100. The measured latitude / longitude data is taken into the tablet PC 241 from the USB port via the signal cable 243. The multi-spectral image shown in FIG. 13 obtained via the data interface device B 115 The pixel / pixel group correspondence table 213, the pixel group / attribute correspondence table 214 for which partial correspondence is undetermined, and the tree information DB 136 are: Tablet pc
The multi-spectral image data device 112, the attribute group data device 113, and the forest information DB device 114 are included in the 221. The multispectral image 150 may be imported together with the multispectral image pixel / pixel group correspondence table 213 and transferred to the multispectral image data device 112 for reference display.

The function of the portable attribute data specifying device 110 will be described below with reference to the flowcharts of FIGS. 22 to 26 related to FIG. 21 and the display and operation of the display screen 242 of the tablet PC 241 and the touch panel. FIG. 21 is a display example of the display screen 242 when the portable attribute data specifying device 110 is used in the state of FIG.
The function of the portable attribute data identification device 110 is processed in block 010 140. The cluster analysis process cannot handle the pixel group / attribute correspondence table 234. There are two functions for confirming or correcting the tree information DB 136 obtained by calculating the tree position from the tree crown. In the following, both will be described together, but the latter tree position correcting function corresponds to the processing block 013 143 tree position calculating process described later.

In FIG. 20, the tablet PC 241 is directed in a target position direction 246 that is the direction of the target position 245 of the target 248 for which an attribute is to be set (target display direction 247). The tablet PC 241 is equipped with an orientation sensor 119 using an electronic compass. The current position 244 is known by the DGPS 116 and the target display direction 247 is known by the orientation sensor 119 and can be reflected on the display screen 242.
You can import the direction of the electronic compass on Android as follows.
sensorManage = (SensorManager) getSystemService (Context.SENSOR_SERVICE);
sensor = sensorManager.getDefaultSemsor (Sensor.TYPE._MAGNETIC_FIELD);
The laser range finder 118 is installed with the horizontal direction fixed to the column 240 or the tablet PC 241 so that it faces the front of the display screen 242. The laser rangefinder 118 is publicly implemented like Leica Disto D8, and when the distance measurement button is pressed, the distance to the target 248 in the target display direction 247 is measured and taken into the tablet PC 241.

FIG. 21 shows a display example corresponding to FIG.
(1) When “current position” of the operation mode switch 253 is selected The current position 244 and target position 245 are displayed on the forest display section 256 of the display screen 242. The target position 245 is a marker that displays a point away from the current position 244 in the target display direction 247 by the distance displayed in the current position data display field 250. The target display direction 247 coincides with the display screen 242 forward direction.

In the forest display unit 256, the region including the target position 245 is determined as the target 248, the corresponding pixel of the multispectral image 150 is obtained from the latitude and longitude of the target position 245, and the multispectral image pixel / pixel group correspondence table is obtained. A corresponding pixel group is obtained from 233, and an attribute (in this case, a tree species code) is obtained from the pixel group / attribute correspondence table 234. On the forest display unit 256, the pixel on the forest display unit 256 corresponding to the pixel on the multispectral image 150 of the pixel group corresponding to the attribute is identified and displayed with the color and pattern as the target 248.
The part other than the target 248 of the forest display section 256 is obtained by searching the multispectral image pixel / pixel group correspondence table 233 from the correspondence between the forest display section 256 pixels and the multispectral image 150 to obtain the corresponding pixel group. An attribute (in this case, a tree species code) is obtained from the / attribute correspondence table 234, and is identified and displayed by color and pattern for each attribute as a non-target 249 on the forest display unit 256.

The tree position is displayed in the forest display section 256. The tree position of the forest display section 256 corresponds to the pixel on the multispectral image 150, and each pixel on the multispectral image 150 corresponds to the latitude and longitude, so the tree information DB 136 is searched and displayed. You can choose. The tree information DB 136 has a field of “information reliability”, and a portable attribute data identification device in the past.
In the case of the position actually measured in 110, it is identified and displayed as “a priori measured tree position 254” (illustrated with a double circle). “A priori measured tree position” is treated as having the highest information reliability. When a tree position is extracted by image processing, it is identified and displayed as a prior image extracted tree position 255 (illustrated by a black circle).

(A) When “ranging” is selected in the ranging operation SW 253, the tree position measurement by the laser range finder 118 is provided because the recognition of the measured position from the tree crown cannot be made zero by image processing. Means. When “Range” is selected, the “Distance” display is highlighted and the display scale at that point is maintained. Next, change the direction of the laser range finder 118 to capture the target with the finder, and press the distance measurement button. The distance to the target is measured, and the target position 245 comes to the position from the top center of the screen in FIG. The scale is automatically converted, and the display on the forest display section 256 is also updated according to the scale.
As a result of the actual measurement, if the prior measured tree position 254 is closest to the prior image extracted tree position 255 (for example, 1 m or less), the prior image extracted tree position 255 is forcibly deleted and replaced with the prior measured tree position 254. .
As shown in FIG. 21, the tree displayed at a position different from the actual tree can be selected for the purpose of erasing by touching the corresponding tree on the forest display section 256 on the screen. Further, the tree position actually measured by the laser range finder 118 is displayed as the currently measured tree position 257 corresponding to the measured latitude and longitude (illustrated by a bold circle). While changing the azimuth of the display screen 242 and the current position 244, it is possible to select the measured tree position and the wrong tree position for the purpose of erasing one after another. When you click “OK” in the distance measurement operation SW 253, the tree selected for the purpose of deletion is the tree information DB
The current measured tree position is registered in the tree information DB 136, and the display of the forest display unit 256 is also updated.

(B) When “ranging” of the ranging operation SW 253 is not selected The distance between the current position 244 and the target position 245 can be set by gesture input of the Android standard function on the forest display section 256 touch panel. The display screen 242 is directed to the target 248 whose attribute is unknown, the current position 244 is moved so that the target position 245 overlaps the target 248 whose attribute is unknown, and the display scale is changed by gesture input as necessary. The scale set by the gesture input is displayed as the distance between the current position 244 and the target position 245 in the current position data display field 250.
As shown in FIG. 21, the tree displayed at a position different from the actual tree can be selected for the purpose of erasing by touching the corresponding tree on the forest display section 256 on the screen. When you click “OK” in the distance measurement operation SW 253, the tree selected for the purpose of deletion is the tree information DB
It is deleted from 136, and the display of the forest display section 256 is also updated.
(2) When “current position” of operation mode SW 253 is not selected The display of forest display section 256 is DGPS
No longer depend on 116 measurement results, and the current position 244 marker is no longer displayed. Forest display section Arbitrary portions of the multispectral image 150 can be displayed at any scale by gesture input on the 256 touch panel. The heading is north above the display screen 242.

The following operations can be performed regardless of the “current position” of the operation mode SW 253 and the “distance” selection status of the distance measurement SW 253.
When the setting of the attribute operation SW 252 is clicked, the attribute of the target 248 surrounding the target position 245 can be set to the attribute highlighted in the attribute display field 251. In the pixel group / attribute correspondence table 234, the attribute of the pixel group whose attribute has not been determined can be determined or changed. All of the pixels corresponding to the pixel group on the display screen 242 have a display corresponding to the input attribute. In this way, all the pixel group attributes whose attributes have not been determined in the pixel group / attribute correspondence table 234 can be determined in the field.

A method for realizing the functions shown in FIG. 21 based on an Android-equipped tablet PC will be described with reference to FIGS. By clicking “in-port” of the operation mode SW 253 on the display screen 242, FIG. 22 (1) is activated, and in processing block 090 260, the multi-spectral image 150, multi-spectral image pixel / pixel group is transmitted from the data interface device B. Correspondence table 233, pixel group / attribute correspondence table
234 and the tree information DB 136 are read and stored in the multispectral image data device 112, the attribute group data device 113, and the forest information DB device 114. In processing block 091 261, the display screen position, scale, and orientation are initialized. Since the process of FIG. 26 (described later) is periodically started, target information is displayed on the display screen 242 according to the initialized parameters.
When the export of the operation mode SW 253 on the display screen 242 is clicked, FIG. 22 (2) is activated, and the processing block 092 262 causes the multispectral image data device 112, the attribute group data device 113, and the forest information DB device. 114 to data interface device B115 Multispectral image 150, Multispectral image pixel / pixel group correspondence table 233, Pixel group / attribute correspondence table 234, Tree information DB
136 is written, and attribute data acquired by the portable attribute data identification device, which is post-processing of the forest information management device fixing unit, is processed by processing block 022 162 of FIG. 12 to perform multispectral image region classification processing of the present invention. 138 is completed.

When the “ranging” of the ranging operation SW 253 in FIG. 21 is clicked and turned ON, FIG. 23 (1) is activated. In processing block 100 270, the operation mode SW 253 highlights “current position” and turns it ON. In processing block 101 271, the operation mode SW 253 “compass” is highlighted and turned ON. At processing block 102 272, the scale currently being set is maintained, and at processing block 103 273, “ranging” of ranging operation SW 253 is highlighted and turned ON. In processing block 104 274, the distance measuring data input / output command issuance processing for waiting for the distance measuring data is performed for the laser distance measuring device 118.
When the distance measurement button of the distance meter is pressed to acquire data, FIG. 23 (2) is activated by data transmission from the distance measurement system, and the distance measurement result data is fetched at processing block 110 280, and at processing block 111 281 Based on the current position measured by the DGPS 116, the azimuth data by the azimuth sensor 119, and the distance measurement result, the target position is calculated in latitude and longitude. In processing block 112 282, the tree information DB 136 is searched to determine whether the tree coordinates are within 1 m, and if present, it is determined to be the same as the existing tree, and the tree position of the tree information DB is measured in processing block 114 284. As the value, the information reliability is set to “a priori actually measured tree position value. If it does not exist, it is determined that the tree is different from the existing tree, and the tree position measured in processing block 113283 is temporarily registered.

When the current position of the operation mode SW 253 is clicked in FIG. 21 to highlight it and turn it ON, the current position by the DGPS 116 is always captured. Click and highlight the compass of the operation mode SW 253 to turn it ON, and take the orientation information that the display screen 242 is oriented from the electronic compass built in the tablet PC. The display screen 242 is displayed when the processing of FIG. 26 is periodically started. The activation cycle is preferably 250 ms or less. In processing block 160 330, it is checked whether or not “current position” is selected by the operation mode SW 253. If it is selected, processing block 162 332 sets the DGPS position information to the current position. If not selected, the current position is determined by the gesture setting value on the display screen 242 forest display section 256 in processing block 161 331. Gesture setting values are captured as shown in FIG.

In processing block 163 333, it is checked whether “compass” is selected in the operation mode SW 253. If it is selected, the processing screen 165 335 sets the screen upper direction direction from the compass measurement value. If not, processing block 164
At 334, set the upper part of the screen to the north.
In processing block 166 336, it is checked whether “ranging” is selected by the ranging operation SW 253. If it is selected, processing block 169 339 reduces the distance between the current position display and the target position display to be distance measurement data. And the display is changed, and the pixel-attribute correspondence diagram is displayed in processing block 170 340 by the set current position, upper screen direction, scale, and pixel-attribute correspondence data.
If not selected, the display scale by the gesture input value is set in the processing block 167 337, and the current position / upward orientation / scale and the pixel-attribute correspondence data in which the pixel-attribute correspondence diagram is set in the processing block 168 338 Is displayed.

Android, which is the operating system of the tablet PC 221, provides a “gesture” that responds to the movement of the finger on the tablet, and FIG. When “ranging” of the ranging operation SW 253 is selected in the processing block 120 290, a tree displayed on the forest display unit 256 is selected by screen touch. If the screen touch indication position is captured at processing block 128 298 and it is determined that the tree is being displayed near at processing block 129 299, processing block 130
In 300, it is registered as an erasure position candidate, and the display color is changed and accepted. If it is determined in processing block 129 299 that the tree is not in the vicinity of the tree being displayed, the process ends without doing anything.
When the “ranging” of the ranging operation SW 253 is not selected in the processing block 120 290, the processing blocks 122 to 125 292 to 295 are operations for moving the display screen center coordinates to the east, west, north, and south. When the “current position” of the operation mode SW 253 in FIG. 21 is not selected, the display screen 242 is not displayed around the current position 244, and the display screen center coordinates are moved east, west, south, and north by “gesture”. The processing blocks 126 to 127 296 to 297 enlarge / reduce the scale of the display screen regardless of the selection of the operation mode SW 253, change the scale display of the current position data display field 250 in FIG. 21, and reduce the scale of the display screen 242. Is reflected in the processing of processing block 167 337.

When “OK” is turned ON by the distance measurement operation SW 253 on the display screen 242, FIG. 24 (1) is activated, and the erasure position candidate (processing block 130 300) is set in the tree position of the tree information DB in the processing block 131 301. Erase from. The registration position candidate (processing block 113 283) is registered in the tree position of the tree information DB.
If the attribute of the target 248 displayed at the target position 245 in FIG. 21 is unknown, the attribute can be set or corrected by operating the attribute display field 251 and the attribute operation SW 252. The attribute setting operation is performed according to the flowcharts of FIGS. 25 (1), (2), and (3). The number on the left side displayed in the attribute display field 251 is the pixel group number, the right side indicates the attribute data (tree species here), and can be scrolled up and down by gesture.

When a gesture is performed on the attribute display field 251, the processing block 145 in FIG.
Processing branches to processing block 146 316 or processing block 149 319 depending on the gesture direction. In processing block 146 316, if the top item of the attribute selection unit is selected, the display item of the attribute selection unit is scrolled up and highlighted in processing block 147 317 to select the top item of the attribute selection unit. If not, processing block 148 318 moves the selection item in the attribute selection section upward to highlight it. Processing block 149 319 is downward in direction but can be discussed similarly.
When attribute data is selected in the attribute display field 251 and “setting” of the attribute operation SW 252 is clicked, the processing of FIG. 25 (2) is started, and the selection item of the attribute selection unit is read in processing block 140 310. In 311, the pixel position on the pixel-attribute correspondence diagram is obtained from the target cursor position on the display screen position, and the attribute of the same group on the pixel-attribute correspondence diagram is determined in processing block 142 312.
When “reset” of the attribute operation SW 252 is clicked in the attribute display field 251, the pixel position on the pixel-attribute correspondence diagram is obtained from the target cursor position on the display screen position in the processing block 143 313, and the pixel-attribute is processed in the processing block 144 314. Reset the attributes of the same group on the corresponding diagram.
Tree information DB 136 manages trees individually, and character information files for each tree
166, image information file 167, audio information file 168, and video / audio information file 169 have link functions. Collect and input these information corresponding to trees in the forest and transfer them to the forest information management device fixing unit 100. I can do it. The forest information management device fixing unit 100 can use these tree information DBs 136 for forest management forest information processing 165 (described later).

This is the end of the description of the processing block 021 161 in FIG. 12, and the processing block 022 162 in the next stage will be described. In processing block 022 162, the correspondence relationship between the integrated pixel group and the attribute data acquired by the portable attribute data specifying device 110 is integrated in the forest information management device fixing unit to determine a representative pixel vector corresponding to the attribute data. The contents of the multi-spectral image pixel / pixel group correspondence table 233 and the pixel group / attribute correspondence table 234 are updated as a result of the processing block 021 161, and this result is used as the data interface device A 105. Then, it returns to the multispectral image data device 102, the attribute group data device 103, and the forest information DB device 104 of the forest information management device fixing unit 100, and determines a representative pixel vector (nuclear vector) corresponding to each attribute data.

In processing block 023 163, the attribute data obtained in the n target regions and the corresponding representative pixel vector (nuclear vector) are applied to the entire forest region 151 or the entire multispectral image 150. Normalize the pixel vector of the entire forest area 151 or the entire multispectral image 150, and then perform an inner product operation between the pixel vector and the attribute data and the corresponding representative pixel vector (nuclear vector-normalized vector) By assigning the attribute data corresponding to the nuclear vector closest to 1.0 above the threshold to the pixel vector, the multispectral image 150 or the forest region 151 is classified by the attribute data.
Fig. 5 Multispectral image region classification process
The description of 138 ends.

Hereinafter, processing block 011 141 in FIG.
Tree Shading Model generation and processing block 013
143 Explain the tree positioning. These three processes are techniques for determining the tree position from the panchromatic image 139 by detecting the crown. There are large tree crown extraction methods: (1) Watershed, (2) Region Growing, and (3) Valley Following. These are all processed using gradients between pixels and belong to differential operations. Therefore, it has a drawback that it is vulnerable to noise or generates noise by itself. The method of the present invention increases the noise resistance and improves the discrimination capability by performing an integral operation (product-sum operation) instead of the differential (difference operation and division) operation in the conventional method.

Processing block 011 141 in FIG. 5 Histogram leveling will be described below with reference to FIGS. Histogram leveling is a type of spatial filter. The histogram leveling filter 363 in FIG. 27 is a new image 107 and the same area as the multispectral image 150 is simultaneously captured at the upper left and HV coordinates (0, The calculation is performed by scanning one pixel at a time from 0) to the lower right and (N-1, M-1) in the HV coordinates. The histogram leveling filter 363 has 2n + 1 pixels in the vertical direction and 2n + 1 pixels in the horizontal direction. It is preferable to set the histogram leveling filter to approximately 50mX50m corresponding to the tree crown size.

28 and 29 show the principle of the histogram leveling filter 363.
The panchromatic image 363 covered by 363 performs processing for maximizing the amount of information (entropy) in the 139 region. The principle and specific implementation method will be described below. Since the size of the leveling filter 363 is set to about twice that of the tree crown to be detected, the meaning of maximizing the amount of information (entropy) in this range is to eliminate information unnecessary for tree crown detection. is there.

28 (1) The continuous original image histogram is a model in which the probability density distribution P (x) 378 of the pixel luminance of the panchromatic image 139 region covered by the histogram leveling filter 363 is shown as a continuous system. (It is actually a discrete system, but starts from a continuous system as the starting point of the discussion.) When the probability density distribution P (x) 378 is given for the maximum luminance value a 373 and the minimum luminance value b 374, the amount of information is Hp. It is known that the following relationship holds when the amount of information for an arbitrary probability density distribution Q (x) 379 defined and different from P (x) is defined by Hq.
Here, consider the transformation from (1) continuous original image histogram to (2) continuous leveled histogram.

The continuous system realizes a continuous leveling histogram by the conversion of (2), and can realize the maximum amount of information shown in (1). Furthermore, as shown in FIG. 28 (2), the information amount Hp in (1) can be increased to log (Xmax−Xmin) by expanding the luminance range from the luminance lower limit value Xmin 376 to the luminance upper limit value Xmax 375.

Based on the above discussion, the correspondence to the discrete system is performed in FIGS. 28 (3) and 28 (4). In actual image processing, the pixel luminance value is expressed as an integer, and therefore, an integer from 0 to 255 to 0 to 4095 is a discrete value. The relational expressions (1) and (2) can be rewritten as follows.
However, in the discrete system, even if the conversion of (5) is performed, (3) and (4) cannot be realized. (Fig. 28 (4) Discrete leveling histogram is not realized.) This is different from the probability density distribution P (x) of the continuous system and the luminance axis X
This is because Pi cannot level (3) even if is operated.

In order to solve this problem, the regions of the panchromatic image 139 corresponding to the histogram leveling filter 363 are resampled in FIGS. 29 (1) to (3). (1) The original image is usually an odd number because the calculation result is written in the center pixel with the same number of pixels in the vertical and horizontal directions. (2) Pixels are divided into four equal parts in a 4 × resampled image, and (3) 16 × resampled images are resampled into 16 equal parts. Resampling is as follows: (1) The pixel luminance area of the original image is converted to a floating point and then a resampling method such as bi-linear or spline is used. (4) With respect to the discrete system original image histogram, (5) a discrete system 4 times resampled image histogram and a discrete system 16 times resampled image histogram (not shown in FIG. 29) are obtained.
A histogram is obtained for the resampled pixels, the luminance value of each pixel is converted by the equation (5), the average value of the oversampled pixels is obtained, and (1) the pixel luminance value corresponding to the original image is determined as a fixed point. Ask for. Finally, integerization corresponding to the number of bits of the pixel is performed, and the central pixel luminance value of (2n + 1) x (2n + 1) of (1) is stored as a result in the panchromatic image result area. The above processing is shown in the processing flow of FIG. FIG. 31 shows an example of an original image, and FIG. 32 shows an image subjected to histogram leveling processing. As the resampling multiple increases, the histogram after histogram leveling in FIG. 32 approaches a uniform distribution, but does not become completely uniform in a discrete system. The optimum value is determined by comparing the processing load and the image quality.

Returning to FIG. 5, processing block 012 142 Tree Shading Model generation processing will be described with reference to FIGS. 33 to 35.
FIG. 33 shows a model of a crown, and conifers and broad-leaved trees are all expressed by spheroids, and tree species and tree sizes all correspond by changing the parameters of the spheroid. As shown in FIG. 33 (4) solar radiation and line-of-sight direction from shooting condition 135, the line-of-sight direction is set as the shooting parameter.
When the solar radiation direction 381 is given, it is possible to obtain a shading diagram viewed from the line-of-sight direction for each of (1) conifer A, (2) conifer B, and (3) broadleaf. (FIG. 33 (b) line-of-sight direction)
Shading is calculated using the well-known Ray
The tracing method may be used. The tree shading model in FIG. 34 is a general tree mathematical model, and an ellipsoidal coordinate system 392 includes x, y, and z axes, and defines a tree model based on a spheroid around the origin. The View coordinate system 393 is a plane perpendicular to the line segment oO indicating the line-of-sight direction 380, and indicates the coordinate system (X, Y axes) of the panchromatic image 139 and the multispectral image 150.

The processing flow of the above processing is shown in FIG.

Next, the tree position calculation processing of the processing block 013 143 in FIG. 5 will be described with reference to FIGS.
Prior to this processing, processing block 010 140 Since the multispectral image 150 has already been classified for each tree species in the cluster analysis process, the panchromatic image 139 with overlapping image regions can be classified according to the tree species. In response, all the tree species are processed sequentially in process block 014 144. Processing block 015
Clearing the tree species area not selected in 145 to zero is intended to eliminate the effects of different tree species. Next, a tree position search process is performed on the selected tree species region of the panchromatic image 139 at process block 016 146. Details will be described below with reference to FIG.

The reason why the Shading Pattern series corresponding to the tree species is selected in the processing block 220 420 is that all tree species are represented by the spheroid parameters a, b, c, and d in FIG. It is expressed by changing the ratio of a, b = c (rotational symmetry on the vertical axis) and d, and the Shqding pattern as seen from the line of sight, for example, is prepared between 2m and 20m in diameter at intervals of 2m. It is to select a pattern file corresponding to the tree species from what was left.
The processing block 221 421 divides the image area into N images because the process of searching for the maximum luminance level point in the area performed in the processing block 223 423 does not have to be performed for the entire screen, and the maximum comparison in the neighborhood is performed. This is because it is only necessary to seek. For example, N = 100 may be used. The sequential processing for each image area in the processing block 222 422 is sequentially performed for each area divided in the processing block 221 421. It should be noted that it is desirable to overlap the maximum crown size in the association part of the region.

Processing blocks 223-227 423-427
Will be described with reference to the processing example of FIG.
Processing block 223 423 finds the maximum brightness balance in the divided area. The maximum luminance point in the garden shown in FIG. 39 (2a) is searched. In the processing block 224 424, the processing of the processing block 225 425 is sequentially performed for each size of the Shading Pattern series of the selected tree type. In processing block 225 425, ± δ from top to bottom and left and right around the maximum luminance point
The position with the highest correlation is searched while shifting pixels. Further, the processing block 224 424 determines the Shading Pattern size having the largest correlation. (FIG. 39 (2a)) This position and size are taken as a tree crown (FIG. 39 (2b)). Processing block 226
If 426 is Shading Pattern maximum correlation value <threshold is No, it means that it is a tree crown instead of noise, so processing block 227
The position and size of the Shading Pattern of the maximum correlation value searched in 427 is registered in the forest information DB as the tree crown position and size, and the corresponding area of the image is zero-cleared. In FIG. 39'3a), the detected crown area is blacked out. If it is determined that the tree crown detected in the processing block 226 426 is noise (Shading Pattern maximum correlation value <threshold), it means that all tree crown extraction has been completed. By repeating the processing blocks 223 to 227 423 to 427, the processing shown in the examples of FIGS. 39 (1) to (12a) proceeds sequentially.

Returning to FIG. 5, the tree position data fusion processing 134 will be described. As described above, since there is an error in the method of obtaining the tree position by extracting the tree crown from the image, the following information is integrated to increase the information reliability.
(1) Temporarily previous panchromatic image 139, tree position calculated from multispectral image 150 (2) Tree position measured by portable attribute data identification device 110 (3) Recent panchromatic image 139, multispectral Tree position calculated using image 150
(4) Information that the tree position has been changed by operations such as logging, thinning, tree planting, etc. (5) Information that the position of the tree has disappeared due to a disaster, such as a mountain collapse, etc. (6) The tree position has been fixed in advance by other means
There are each term.

Tree position data fusion processing
134 is an information reliability management process of the processing block 019 149 via the tree information DB 136, a tree position determination process of the processing block 013 143, and a portable attribute data specifying device
110 is a process of integrating the obtained tree positions. The tree position calculation process of process block 013 143 is further illustrated in FIG. In process block 017 147, operation information 120 is captured, and in process block 018 148, disaster information 121 is captured. This information only needs to include tree position information for use in the information reliability management process of processing block 019 149. Information that can grasp the area where thinning, logging, and afforestation were performed by latitude and longitude is possible. If so, a drawing or the like in which the position of the timber after the operation is found is preferable, but the tree position can be calculated by image processing as long as the fact and the scope of the execution are known. Disaster Information
For 121, information that can grasp the area by latitude and longitude is desirable, but it can be calculated from the image even if there is a disaster and the area is roughly specified.

The information reliability management process of process block 019 149 will be further described with reference to the process flow of FIG. 38 and the state transition diagram of FIG.
This part is a part that forms the basis of the present invention for managing the position in units of trees, and enables tree information with different information sources to be integrated.
Tree position information has different reliability depending on the information source, so it is managed in the following four stages.
(1) Information reliability “None” Indicates that there is no position information.
(2) Information reliability “So-So” Indicates that the position information is “OK”.
Specifically, the reliability status of the location information extracted from the image
Indicates.
(3) Information reliability "Questionable"
"Indicates location information is" suspicious ".
The extracted location information is inconsistent and indicates a suspicious state.
(4) Information reliability "Decisive" Indicates that the position information is in a "certain" state and is a measured value.

FIG. 40 is a state transition diagram showing an algorithm for integrated management of tree position information in accordance with the description of the automaton. Legend of how to write
Shown in 431. “A” and “B” surrounded by circles indicate “states”, and arrows indicate transitions between the states. “X / Y” means that Y is performed when the condition X is satisfied. The meaning of the legend 431 is that “when the condition“ X ”is satisfied, the one in the state“ A ”performs“ Y ”and transitions to the state“ B ”” ”.
The information reliability “None” 430 in FIG.
(1) When image analysis is performed, the tree position is updated and updated to shift to the “So-So” state. (2) When the field data is obtained by actual measurement, the tree position is updated to the “Decisive” state. Transition.

The information reliability “So-So” 431 in FIG. 40 indicates that the position information is “OK”.
(1) If the image analysis result coincides with the previous time, the tree position is maintained and the "So-So" state remains.
(2) When there is the operation information 120 or the disaster information 121 and there is detailed data regarding the tree position, the tree position is updated with the detailed data, and the state shifts to the “Decisive” state.
(3) When the field data is obtained by actual measurement, the tree position is updated and the state shifts to the “Decisive” state.
(4) If there is operation information 120 or disaster information 121 and there is no detailed data regarding the tree position, the tree position is maintained and the state shifts to the “Questionable” state.
(5) If the image analysis result does not match the previous result, the tree position is updated and the state shifts to the “Questionable” state.

Information reliability "Questable" in Fig. 40
432 is the position information is "suspicious"
(1) If the image analysis result coincides with the previous time, the tree position is held and the “Questionable” state is held.
(2) When there is the operation information 120 or the disaster information 121 and there is detailed data regarding the tree position, the tree position is updated with the detailed data, and the state shifts to the “Decisive” state.
(3) When the field data is obtained by actual measurement, the tree position is updated and the state shifts to the “Decisive” state.

The information reliability “Decisive” 433 in FIG. 40 indicates that the position information is “certain”.
(1) When field data is obtained by actual measurement, the tree position is updated and the “Decisive” state is maintained.
(2) If there is no operational information 120 and disaster information 121, the tree position is maintained regardless of the result of image analysis, and the "Decisive" state is maintained.
(3) If the image analysis result is inconsistent with the previous time, there is operation information 120 or disaster information 121, and there is no detailed data regarding the tree position, the tree position is updated and the state shifts to the “Questionable” state.

According to the above, even if the same calculation algorithm is used for the same image data source, the ability to recognize the crown and the tree position differs before and after thinning. The tree position can be determined after thinning to improve recognition accuracy. Further, since it has been found that there is an error in the recognition of the tree position from the image, it is possible to make the position information with doubtful reliability highly reliable by the portable attribute data specifying device 110.
Although the process of the state transition diagram of FIG. 40 has been described above, in the processing flow of FIG. 38, if there is a priori information in the processing block 210 400, if there is already measured or calculated information, Processing may be performed by branching in the current state in block 211 401 and in the processing blocks 212 to 215 402 to 405.
Finally, returning to FIG. 5, forest management forest information processing
165 will be described. The content of the processing is shown in Fig. 6 (3). Processing block 045 185 Calculation of secondary information of trees, monitoring of aging of the tree information and statistical processing that require statistical processing, calculation of breast height diameter, accumulation, yield ratio number and tree age Update. Methods for calculating breast height diameter, accumulation and yield ratio are well known to those skilled in the art.
The forest feature extraction processing in processing block 046 186 uses the tree information DB 136 and DEM 164 for each tree to search for and analyze the distribution status and growth conditions of specific vegetation, and contributes to commercialization of mountain products with high added value. Further, processing block 047 187 provides forest management support including investment efficiency evaluation.

  The forest information management apparatus of the present invention can manage the forest situation in detail by latitude and longitude. For tree species, unknown vegetation that has not been recognized in advance can be detected, so it is possible to manage rare plants and manage and collect high value-added “Yama no Sachi”. It is also possible to detect that illegal plants are cultivated in the mountains.

It is a figure which shows the whole structure of the forest information management apparatus of this invention. It is an example of the Hayashi Kobayashi group diagram. This is an example of a broadleaf tree canopy. This is a copy of an empirical verification paper on the ability of a tree to detect trees. It is a figure which shows the whole structure of the information flow of the forest information management apparatus of this invention. It is a figure which shows the whole structure of the information processing of the forest information management apparatus of this invention. It is a figure which shows the structure of tree information DB (database) of the forest information management apparatus of this invention. It is a figure which shows the principle of the multispectral image area classification | category process of the forest information management apparatus of this invention. It is a figure which shows the correlation calculation area | region setting method of the multispectral image of the multispectral image area classification | category process of the forest information management apparatus of this invention. It is a figure which shows the method of the correlation calculation between the pixels of the multispectral image of the multispectral image area classification | category process of the forest information management apparatus of this invention. It is a figure which shows the concept of the integration process of the pixel group for every object area | region in the multispectral image of the multispectral image area classification | category process of the forest information management apparatus of this invention. It is a figure which shows the whole structure of the multispectral image area | region classification | category process of the forest information management apparatus of this invention. 3 is a flowchart showing the contents of a processing block 020. 3 is a flowchart showing the contents of a processing block 030. 10 is a flowchart showing the contents of a processing block 032. 10 is a flowchart showing the contents of a processing block 051. 10 is a flowchart showing the contents of a processing block 052. 10 is a flowchart showing the contents of a processing block 054. It is a figure which shows the structure of a multispectral image pixel / pixel group correspondence table and a pixel group / attribute correspondence table. It is a figure which shows operation | movement of the portable attribute data identification apparatus 110 of this invention. It is an example of a display screen of the portable attribute data identification device 110 of the present invention. It is a flowchart which shows the process corresponding to the display screen of the portable attribute data specific device 110 of this invention. It is a flowchart which shows the process corresponding to the display screen of the portable attribute data specific device 110 of this invention. It is a flowchart which shows the process corresponding to the display screen of the portable attribute data specific device 110 of this invention. It is a flowchart which shows the process corresponding to the display screen of the portable attribute data specific device 110 of this invention. It is a flowchart which shows the process corresponding to the display screen of the portable attribute data specific device 110 of this invention. FIG. 10 is a diagram showing an operation of histogram leveling in a processing block 002. FIG. 3 is a conceptual diagram illustrating a principle of histogram leveling in a processing block 002. FIG. 3 is a conceptual diagram 2 showing the principle of histogram leveling in a processing block 002. 10 is a flowchart showing the contents of histogram leveling in a processing block 011. It is an example of image data before the histogram leveling process of the processing block 002. It is an example of image data after the histogram leveling process of the processing block 002. FIG. 10 is a diagram illustrating crown modeling in the generation of a tree Shading Model in a processing block 003. FIG. 10 is a diagram illustrating a mathematical expression modeling method for a tree crown in generating a tree Shading Model in a processing block 003; 14 is a flowchart showing the contents of histogram leveling in a processing block 013. 3 is a flowchart showing the contents of a processing block 012. 4 is a flowchart showing the contents of a processing block 014. 10 is a flowchart showing the contents of a processing block 015. It is a figure which shows the example of a tree crown recognition process. It is a figure which shows the state transition of an information reliability management process.

100 Forest information management device fixing part
101 Arithmetic unit A
102 Multispectral image data device
103 Attribute group data device
104 Forest information database system
105 Data interface device A
106 Operator
107 New images
108 Standard GCP
109 Operation information data device
110 Portable attribute data identification device
111 Arithmetic unit B
112 Multispectral image data device
113 attribute group data device
114 Forest information database system
115 Data interface device B
116 DGPS
117 Display and data input device
118 laser rangefinder
119 Direction sensor
120 Operational information
121 Disaster information
122 Crown 1
123 to 125 Crown 2-1 to Crown 2-3
126 DEM data device
127 Forest management information device
130 Processing block 001
132 Hayashi Small Team
133 Hayashi Kodama branch "5-ro-39"
134 Tree position data fusion processing
135 Shooting conditions
136 Tree Information DB
137 Forest management information
138 Multispectral Image Region Classification Processing
139 panchromatic images
140 to 149 Processing block 010 to Processing block 019
150 multispectral image
151 Forest area
152 to 158 Target area 1 to target area i to target area n
160 to 163 Processing block 020 to Processing block 023
164 DEM
165 Forest management Forest information processing
166 Character information file

167 Image information file

168 Audio information file
168 Video information file
170 to 176 Processing block 030 to Processing block 036
180 to 183 Processing block 040 to Processing block 043
185 to 187 Processing block 045 to Processing block 047
190 to 194 Processing block 050 to Processing block 054
200 to 203 Processing block 060 to Processing block 063
210 to 216 Processing block 070 to Processing block 076
220 to 226 Processing block 080 to Processing block 086
230 H coordinate axis
231 V coordinate axis
232 pixels kl
233 Multispectral image Pixel / pixel group correspondence table
234 Pixel group / attribute correspondence table
240 props
241 tablet PC
242 display screen
243 signal cable
244 Current position
245 Target position
246 Target position direction
247 Target display direction
248 goals
249 Non-target
250 Current position data display column
251 Attribute display field
252 Attribute operation SW
253 Distance control SW
254 Location of trees measured a priori
255 Prior Image Extraction Tree Position
256 Forest display
257 Current measured tree position
260 to 262 Processing block 090 to Processing block 092
270 to 274 Processing block 100 to Processing block 104
280 to 284 Processing block 110 to Processing block 114
290 to 301 Processing block 120 to Processing block 131
310 to 321 Processing block 140 to Processing block 151
330 to 340 Processing block 160 to Processing block 170
350 to 356 Processing block 180 to Processing block 186
360 H coordinate axis
361 V coordinate axis
362 pixels kl 362
363 Histogram leveling filter
370 Frequency axis Y
371 Luminance axis X
372 Leveling frequency
373 Maximum brightness a
374 Minimum brightness b
375 Upper limit of brightness Xmax
376 Lower brightness limit Xmin
377 Luminance value x
378 probability density distribution P (x)
379 probability density distribution Q (x)
380 Luminance value xe
381 Frequency distribution Pi
382 Frequency distribution Qi
383 Luminance value xi
384 luminance value xei
390 Tree ellipsoid model
391 Tree surface position
392 Ellipsoidal coordinate system
393 View coordinate system
380 Gaze direction
381 Solar radiation direction
400 to 405 Processing block 200 to Processing block 205
410 to 417 Processing block 210 to Processing block 217
420 to 427 Processing block 220 to Processing block 227
430 Information reliability "None"
431 Information reliability "So-So"
432 Information Reliability "Questionable
"
433 Information reliability "Decisive"
434 Legend

Claims (3)

Information on forests with different measurement timing and measurement methods (forest information), including latitude and longitude, including tree species, forest age, municipality identification information, forest group identification information, small group identification information, branch number identification information A forest information management device characterized by identifying information (tree information) corresponding to individual trees constituting the forest, managing and updating the forest information,
As a method for measuring individual tree information constituting the forest, it is composed of (1) pixels of multispectral multispectral image of aerial image of satellite image or satellite image calibrated by GCP (Grand Control Point). The region of the image is classified by the crown region and the tree species by the pixel of the image, and the panchromatic image of the ortho image is subjected to the operation of leveling the histogram locally with respect to the classified image, and the solar radiation direction. By detecting the crown by correlation with the shading model of the tree including correspondence to the shooting direction, obtaining the position of each tree based on the latitude and longitude, obtaining the size of the crown, and obtaining the tree species, the forest A method characterized by adding individual tree information constituting the information to the forest information;
And as another method for measuring tree information,
(2) Precision positioning device including DGPS (Differential Global Positioning System), image display device, data input device, arithmetic device, electronic compass, laser distance meter, and image data, tree species region classification data, tree position Using the portable tree information identification device composed of components including the database device that stores the data including latitude and longitude data, the vicinity of the ground where the actual object corresponding to the pixel included in the classified area can be directly visually confirmed The portable tree information identification device is installed in, the latitude and longitude of the installation position are measured by the precision position locator, the direction of the target object is measured by an electronic compass, and the distance from the target object is visually or laser measured. To measure the position of individual trees based on the latitude and longitude,
An image including multispectral image data corresponding to the latitude and longitude inside the database device and region classification data of the multispectral image is retrieved and displayed on the image display device, and the pixels included in the classified region and the actual target object are displayed. Confirming the correspondence with individual tree information including tree species, inputting the confirmation result from the data input device, obtaining the tree species, obtaining the size of the crown, and making it forest information,
A forest information management apparatus using a method for measuring tree information including (1) and (2) . 2. An image according to claim 1, wherein the direction for visually recognizing the screen of the image display device of the portable attribute data specifying device is the same as the direction for visually recognizing the actual object corresponding to the pixels included in the classified area. An orientation detector is installed in the display device, the display screen is rotated in response to the detection result of the orientation detector, and the upper part of the screen is always in the direction of visually recognizing the actual object corresponding to the pixels included in the classified area. To be identical,
Furthermore, the laser rangefinder is directed toward the target object at the top of the screen to measure the distance to the target object, or the scale of the display screen is manually changed and the distance to the target object is visually measured. Portable tree information identification device characterized by In Claim 1, the reliability of the tree information including the position of the tree having a different measurement time and measurement method is evaluated and updated by using the position information including the latitude and longitude specified for each individual tree as a reference. Forest information management device characterized by