A method for regularly inspecting the growth status of planted trees

Abstract

A new method for regularly inspecting the planting and growth status of trees differs from mainstream remote sensing image technology in that it uses sensors within close range of the trees to acquire images of individual trees and transmits the captured images to a remote server via wireless signals. The remote server receives the images and analyzes the images of each tree in a certain area to identify trees with problems and provides the corresponding tree numbers back to the inspectors, greatly reducing the workload of the inspectors.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of forestry, computer science, and pattern recognition technology, and in particular, relates to a method for regularly inspecting the planting and growth status of trees. Background Technology

[0002] Monitoring the planting and growth status of trees is a crucial part of forestry ecological analysis. With the development of information technology, image and video methods have become important tools for monitoring tree planting and growth, and are widely used in production practices. These methods are applied to various forestry management applications, including forest resource monitoring, afforestation and logging, forest stand boundary identification, biological characteristic assessment, biodiversity conservation, and forest fire resource loss monitoring. In particular, accurate and rapid monitoring of seedling growth and disease conditions is especially important during seedling cultivation and research.

[0003] Monitoring the planting and growth status of trees is a key requirement for forest resource monitoring. Currently, the mainstream monitoring method uses remote sensing imagery, extracting color and outline features within a specific area from high-resolution images. However, while remote sensing images can extract features over a large area, they are not precise enough for extracting information about individual trees, such as detecting pests and diseases in a particular tree. Furthermore, remote sensing images primarily focus on leaf density and color, failing to monitor pest infestations on the trunk. Currently, manual visual inspection is often used to supplement individual tree detection, but this is labor-intensive, resource-intensive, and dependent on the expertise of personnel, resulting in high training costs. Therefore, there is an urgent need for an automated method to monitor tree planting and growth status with high accuracy for individual trees, enabling timely detection of trees exhibiting growth problems.

[0004] Existing technologies also employ image processing techniques such as neural networks for tree detection, but most focus on single macroscopic aspects like tree outline, size, and height, lacking suitable, specialized algorithms for detecting comprehensive conditions such as pests, diseases, and growth status. Furthermore, directly applying neural network models from other fields yields poor results and cannot be industrially scaled up. Summary of the Invention

[0005] This invention creatively proposes a method for regularly inspecting the planting and growth status of trees. Unlike mainstream remote sensing imagery monitoring methods, it uses sensors within close range of the trees to acquire images of individual trees and transmits the captured images to a remote server via wireless signals. The remote server receives the images and analyzes the images of each tree within a certain area to identify trees with problems, and then feeds back the corresponding tree numbers to the inspection personnel, greatly reducing the workload of the inspection personnel. Furthermore, the image analysis method based on individual trees proposed in this invention is also a significant innovation.

[0006] A method for regularly inspecting the growth status of planted trees.

[0007] Step 1: Using image sensors deployed around the tree, images of the tree trunk and branches are acquired. These images are then transmitted wirelessly to a remote server. The server combines the images from eight sensors around the same tree into a single dataset, denoted as [data set name missing]. ,in This indicates a location marker; the superscript number in the set indicates the order in which the trees are surrounded. These represent images taken facing the tree trunk and images taken facing the branches, respectively.

[0008] Step 2: Identify the trunk area and check for pests and diseases.

[0009] The original image of the photo taken facing the tree trunk. With three different edge response operators , , , , , Perform convolution operations to obtain the edge response map. :

[0010]

[0011]

[0012]

[0013]

[0014] Where max represents the maximum value among the calculation results of several response operators, and the symbol... This means that the image is binarized according to the threshold, that is, divided into responding and non-responding, to obtain the edge response map; the centroid of each edge response map is taken, and the set of pixels through which the line connecting the responding edge and the centroid passes is the trunk region.

[0015] Based on the obtained tree trunk area, K types of pest and disease image templates are used to scan the tree trunk area to obtain the probability values ​​of pests and diseases present in different areas, and these probabilities are recorded as follows: This indicates that the probability of matching the image template of the k-th type of pest or disease is greater than the threshold. The number of local regions, and:

[0016]

[0017] , These represent the categories that have the most matches with all local regions within the tree area;

[0018] like If so, the tree is marked as potentially having pests or diseases in its trunk;

[0019] Step 3: Identify the branch areas and check the foliage coverage.

[0020] Take images facing the tree branches Extract consecutive sub-images from each image, with each sub-image having a size of [size missing]. Perform feature transformations on each subgraph:

[0021]

[0022]

[0023] In the above four formulas This represents a sub-image of an image taken towards a tree branch. Represents the pixel coordinates in the sub-image;

[0024] The eigenvalues ​​of the subgraph are calculated using the four formulas described above. This is used to represent the local features of a subgraph; the local features are trained using training samples from the leaf subgraph, and a binary classifier is used to classify a subgraph of an image taken facing a tree branch. The system determines whether a region is a leafy area, and then calculates the leaf coverage on the branches.

[0025] Step 4: Analyze the problem trees based on the judgment results of the trunk and branches, and send their numbers to the user terminal.

[0026] In step 3 above, the leaf coverage area of ​​the image taken towards the tree branch is obtained based on the discrimination result of each sub-image, and its coverage is calculated based on the ratio of the number of covered pixels to the total number of pixels in the image.

[0027] In step 1 above, image sensors are placed on the ground around the tree. There are four locations for the image sensors, with each location opposite to the others and the lines connecting them being orthogonal. Two sensors are placed at each location, one of which is pointing towards the trunk and the other towards the sky and towards the branches.

[0028] In step 1, each sensor takes an image at regular intervals.

[0029] Take one image each morning and afternoon during periods of ample sunlight.

[0030] In step 2, the matching probability The calculation method is as follows:

[0031]

[0032] in , They represent , The median coordinate is The pixel value.

[0033] An inspection system implementing the above method includes a sensor terminal, a server, and a user terminal.

[0034] The sensing terminal includes an image sensor facing the branch and an image sensor facing the trunk, a base and a bracket for supporting the sensors.

[0035] The sensing terminal also includes a wireless transmission device for transmitting images to a server.

[0036] The user terminal is used to receive tree numbers and prompt inspectors to conduct manual inspections of the corresponding trees.

[0037] The inventive points and technical effects of this invention:

[0038] 1. This invention proposes a method for regular inspection of the planting and growth status of trees. Unlike the mainstream remote sensing image technology monitoring method, it uses image sensors placed on the ground around the trees to simultaneously collect and process images of the tree trunk and branches, and combines the two to identify trees with problems. The corresponding tree numbers are then fed back to the inspection personnel, which greatly reduces the workload of the inspection personnel.

[0039] 2. A specialized image processing method for identifying tree trunk regions was designed. Utilizing a special edge operator and an edge correspondence map acquisition algorithm, and through centroid correlation, the tree trunk region can be accurately identified, providing an image basis for pest and disease detection. The pest and disease detection algorithm was optimized to constrain potential errors from the aforementioned detection methods to the greatest extent possible, particularly preventing misidentification of textures as wormholes when the tree texture is too complex. Thus, accurate detection results can be obtained with a computationally significantly lower cost than neural network models.

[0040] 3. Unlike general leaf cover detection, a unique feature value for sub-images is proposed. The extraction method makes leaf identification more accurate and faster than general neural network algorithms and image binarization algorithms.

[0041] 4. A comprehensive judgment method based on the fruiting of the trunk and branches is proposed, which is more suitable for tree inspection during the seedling stage. Attached Figure Description

[0042] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, are not intended to limit the scope of the invention. In the drawings:

[0043] Figure 1 This is the main view of the image sensor deployment.

[0044] Figure 2 This is a top-down view of the image sensor deployment.

[0045] Figure 3 This is a schematic diagram of the trunk region formed based on the center of gravity and edge response. Detailed Implementation

[0046] Normally, assessing tree growth includes checking for pests and diseases, as well as foliage coverage—essential tasks for routine inspections. However, current manual inspections are extremely labor-intensive, and only a small number of trees are identified as having problems, resulting in a significant waste of time. This is particularly problematic for large, important forest areas or nursery areas, where timely detection of abnormal tree growth can prevent substantial economic losses. This is because certain pests and diseases spread very rapidly during the seedling stage; failure to detect them promptly, or the resulting reduction in foliage coverage, can lead to similar problems affecting large areas of the forest. The only solution to this problem would be to increase the frequency of manual inspections, but this is clearly impractical.

[0047] Therefore, this invention proposes a wireless inspection system and a method for its implementation.

[0048] The inspection system includes: sensor terminals, server and user terminals.

[0049] The sensing terminals include image sensors 1 facing the tree branches and image sensors 2 facing the tree trunk, a base 3 and a bracket 4 for supporting the sensors. Four sensing terminals are deployed around each tree 5 to acquire images from four directions, providing a more comprehensive view. Understandably, the number of deployments can be appropriately reduced to lower costs.

[0050] The sensing terminal also includes a wireless transmission device for transmitting images to a server.

[0051] The server receives images, processes and identifies them, determines the growth status of trees, and sends the numbers of trees with potential growth problems to the user terminal.

[0052] The user terminal is used to receive tree numbers and prompt inspectors to conduct manual inspections of the corresponding trees.

[0053] The specific implementation methods include:

[0054] Step 1 Tree image acquisition, transmission, and ensemble construction based on image sensors

[0055] Image sensors deployed around the trees capture images of the tree trunks and branches. These images are then transmitted wirelessly to a remote server. The image aggregation module on the server collects all images captured by the sensors around the same tree and passes them to the next step for analysis.

[0056] S1.1 A method for acquiring images of tree trunks and branches using image sensors arranged on the ground around a tree. Four image sensors are arranged around the tree, with each pair of sensors facing each other and their connecting lines orthogonal, thus optimizing the shooting range of all sensors. Two sensors are arranged at each point, one shooting towards the tree trunk and the other towards the sky shooting towards the branches. The shooting direction of each sensor is pre-set and known. Figure 1 ).

[0057] Under the existing technology, the above-mentioned image sensor can integrate the wireless signal transmission function of a mobile communication chip to transmit the acquired images to a specific remote server; as an optional but not limited method, the above-mentioned image sensor and its associated chip can be powered by underground cables or batteries.

[0058] S1.2 Each sensor takes an image at regular intervals. As a recommended configuration, one image should be taken each morning and afternoon when sunlight is abundant. The shooting time is preset by the patrol personnel, with sunny weather and sufficient light being preferred. Shooting by all sensors in the area is carried out simultaneously.

[0059] After each shooting session, all images captured by the sensors are transmitted wirelessly to a remote server. The remote server is configured with an image aggregation module, which combines images captured by eight sensors around the same tree into a set, denoted as [image set name].

[0060]

[0061] in This indicates a location marker; the superscript number in the set indicates the order in which the trees are surrounded. These represent images taken from the tree trunk and images taken from the branches, respectively.

[0062] Step 2 Identify the tree trunk area and check for pests and diseases.

[0063] The tree trunk area is detected from images taken facing the tree trunk, and the presence of pest and disease characteristics in the tree trunk area is detected.

[0064] S2.1 In the specific environment described in this invention, the image sensor may be partially obscured by the leaves of ground plants. In order to detect the tree trunk area under partial obscuration conditions, a tree trunk area detection method robust to obscuration is proposed.

[0065] The response operator for the edge of the tree region is defined as follows:

[0066]

[0067] in , There are two square formations. , These represent the elements of the matrix's rows and columns, respectively. It is the element index, and:

[0068]

[0069]

[0070] in , express , The number of elements in each row (column). .

[0071] Furthermore, define:

[0072]

[0073] , There are two square formations. . , These represent the elements of the matrix's rows and columns, respectively. It is the element index, and:

[0074]

[0075]

[0076] The above double diagonal lines This represents the integer division symbol. For example, 3 / / 2=1, 4 / / 2=2, 5 / / 2=2.

[0077] Furthermore, define:

[0078]

[0079] , There are two square formations. . , These represent the elements of the matrix's rows and columns, respectively. It is the element index, and:

[0080]

[0081]

[0082] The above , , , , , These represent edge response operators at three different scales, used to adapt the trees to their relative size with respect to the image and to remove local noise.

[0083] The original image of the photo taken facing the tree trunk. Perform a convolution operation with the aforementioned edge response operator to obtain the edge response map. :

[0084]

[0085]

[0086]

[0087]

[0088] Where max represents the maximum value among the calculation results of several response operators, and the symbol... This indicates that image binarization is performed according to a threshold, i.e., it is divided into those with response and those without, to obtain edge response maps; the centroid of each edge response map is then taken:

[0089]

[0090]

[0091]

[0092]

[0093] in , , , They represent

[0094] The position of the center of gravity.

[0095] In the original image, the set of pixels traversed by the line connecting the responding edge and the centroid constitutes the trunk region. Figure 3 ).

[0096] S2.2 Based on the tree trunk area obtained in S2.1, scan the tree trunk area according to the pest and disease image template group to obtain the probability value of the presence of pests and diseases in different areas.

[0097] Let the image template group of pests and diseases be... A partial image of a tree trunk suffering from pests and diseases is denoted as:

[0098]

[0099] Within the tree trunk area obtained in S2.1, select equal-sized local regions and calculate the matching probability with the pest and disease image template group. Repeat this process until all tree trunk areas have been scanned. The number of scanned regions is... The local area scanned each time is denoted as:

[0100]

[0101] Matching probability The calculation method is as follows:

[0102]

[0103] in , They represent , The median coordinate is The pixel value.

[0104] remember This indicates that the probability of matching the image template of the k-th type of pest or disease is greater than the threshold. The number of local regions, and:

[0105]

[0106] , These represent the categories that match the most local regions within the tree area.

[0107] if:

[0108]

[0109] The tree is then marked as potentially infested with pests or diseases. These criteria were optimized through extensive experimentation and can constrain potential errors from the aforementioned detection methods to the greatest extent possible. In particular, they prevent misidentification of complex tree textures as wormholes, thus improving detection accuracy. Experiments show that this detection accuracy is 92%, compared to 80% using the ResNet neural network method under similar conditions, and 54% using general image processing methods (such as binarization).

[0110] Step 3 Identify branch areas and measure leaf coverage.

[0111] The tree branch and leaf areas are detected in images taken facing the sky to determine the leaf coverage.

[0112] Take images facing the tree branches Extract consecutive sub-images from each image, with each sub-image having a size of [size missing]. As the preferred choice .

[0113] Perform feature transformations on each subgraph:

[0114]

[0115]

[0116] In the above four formulas This represents a sub-image of an image taken towards a tree branch. This represents the pixel coordinates in the subimage. The eigenvalues ​​of the subimage are calculated using the four formulas described above. The is used to represent the local features of the subimage. The subimage feature values ​​mentioned above reflect the texture features of the leaf parts in different directions, which improves the robustness of the features compared with conventional pixel-based template matching methods.

[0117] The aforementioned local features are trained using training samples from leaf sub-images, and a binary classifier is used to classify a sub-image of an image taken facing a tree branch. The system performs a discrimination test to determine whether the image is a leafy area. Further, based on the discrimination results of each sub-image, it obtains the leaf-covered area of ​​the image taken facing the tree branch, and calculates its coverage based on the ratio of the number of covered pixels to the total number of pixels in the image.

[0118] Step 4 The planting and growth status of trees can be determined by analyzing the pest and disease status of the trunk area in images taken facing the trunk and the leaf coverage in images taken facing the branches.

[0119] Based on step 2, the image taken facing the tree trunk shows the status of pests and diseases in the trunk area. Based on step 3, the image taken facing the branches shows the leaf coverage. The planting and growth status of the trees can then be further determined.

[0120] Based on the pest and disease status of the tree trunk area obtained from the images taken facing the tree trunk in step 2, each image is marked as 0 or 1; a total of 4 images facing the tree trunk are obtained for a single tree. Their corresponding markers are denoted as:

[0121]

[0122] Furthermore, define:

[0123]

[0124] Based on the leaf coverage obtained in step 3 for the images taken facing the tree branches, a percentage value between 0 and 1 is obtained for each image taken facing the tree branches; a total of 4 images facing the tree branches are obtained for a single tree. Their coverage values ​​are denoted as:

[0125]

[0126] Further calculations:

[0127]

[0128] For all trees within the monitoring area, calculate the corresponding... The distribution of is as follows:

[0129]

[0130] This indicates the total number of trees. , They represent The distribution mean and standard deviation.

[0131] For the nth tree, if its corresponding And its corresponding In overall distribution twice the standard deviation In addition, if the trees are found to have growth problems, the patrol personnel are advised to conduct further manual inspections. These judgment criteria are more suitable for forest patrols during the seedling stage, enabling accurate and rapid identification of problematic trees.

[0132] The table below shows the detection accuracy of the method of the present invention for automatically inspecting the condition of trees with problems. It can be seen that the method of the present invention is applicable to the detection of the planting and growth status of trees over a large area, and can accurately locate the problem of a single tree, greatly reducing the workload of manual inspection.

[0133]

[0134] Those skilled in the art will recognize that, although numerous exemplary embodiments of the invention have been shown and described in detail herein, many other variations or modifications conforming to the principles of the invention can be directly determined or derived from the disclosure of the invention without departing from its spirit and scope. Therefore, the scope of the invention should be understood and construed as covering all such other variations or modifications.

Claims

1. A method for regularly inspecting the growth status of planted trees, characterized in that: Step 1: Using image sensors deployed around the tree, images of the tree trunk and branches are acquired. These images are then transmitted wirelessly to a remote server. The server combines the images from eight sensors around the same tree into a single dataset, denoted as [data set name missing]. ,in This indicates a location marker; the superscript number in the set indicates the order in which the trees are surrounded. These represent images taken facing the tree trunk and images taken facing the branches, respectively. Step 2: Identify the trunk area and check for pests and diseases. The original image of the photo taken facing the tree trunk. With three different edge response operators , , , , , Perform convolution operations to obtain the edge response map. : ； ； ； ； Where max represents the maximum value among the calculation results of several response operators, and the symbol... This means that the image is binarized according to the threshold, that is, divided into responding and non-responding, to obtain the edge response map; the centroid of each edge response map is taken, and the set of pixels through which the line connecting the responding edge and the centroid passes is the trunk region. Based on the obtained tree trunk area, K types of pest and disease image templates are used to scan the tree trunk area to obtain the probability values ​​of pests and diseases present in different areas, and these probabilities are recorded as follows: This indicates that the probability of matching the image template of the k-th type of pest or disease is greater than the threshold. The number of local regions, and: ； This represents the category that matches the most local regions within the tree area; like If so, the tree is marked as potentially having pests or diseases in its trunk; Step 3: Identify the branch areas and check the foliage coverage. Take images facing the tree branches Extract consecutive sub-images from each image, with each sub-image having a size of [size missing]. Perform feature transformations on each subgraph: ； ； In the above four formulas This represents a sub-image of an image taken towards a tree branch. Represents the pixel coordinates in the sub-image; The eigenvalues ​​of the subgraph are calculated using the four formulas described above. This is used to represent the local features of a subgraph; the local features are trained using training samples from the leaf subgraph, and a binary classifier is used to classify a subgraph of an image taken facing a tree branch. The system determines whether a region is a leafy area, and then calculates the leaf coverage on the branches. Step 4: Analyze the problem trees based on the judgment results of the trunk and branches, and send their numbers to the user terminal.

2. The method for regularly inspecting the planting and growth status of trees as described in claim 1, characterized in that: In step 3 above, the leaf coverage area of ​​the image taken towards the tree branch is obtained based on the discrimination result of each sub-image, and its coverage is calculated based on the ratio of the number of covered pixels to the total number of pixels in the image.

3. The method for regularly inspecting the planting and growth status of trees as described in claim 1, characterized in that: In step 1 above, image sensors are placed on the ground around the tree. There are four locations for the image sensors, with each location opposite to the others and the lines connecting them being orthogonal. Two sensors are placed at each location, one of which is pointing towards the trunk and the other towards the sky and towards the branches.

4. The method for regularly inspecting the planting and growth status of trees as described in claim 1, characterized in that: In step 1, each sensor takes an image at regular intervals.

5. The method for regularly inspecting the planting and growth status of trees as described in claim 4, characterized in that: Take one image each morning and afternoon during periods of ample sunlight.

6. The method for regularly inspecting the planting and growth status of trees as described in claim 1, characterized in that: In step 2, the matching probability The calculation method is as follows: ； in , They represent , The median coordinate is The pixel value.

7. An inspection system for implementing a timed inspection method for the planting and growth status of trees as described in any one of claims 1-6, characterized in that: This includes sensor terminals, servers, and user terminals.

8. The inspection system as described in claim 7, characterized in that: The sensing terminal includes an image sensor facing the branch and an image sensor facing the trunk, a base and a bracket for supporting the sensors.

9. The inspection system as described in claim 7, characterized in that: The sensing terminal also includes a wireless transmission device for transmitting images to a server.

10. The inspection system as described in claim 7, characterized in that: The user terminal is used to receive tree numbers and prompt inspectors to conduct manual inspections of the corresponding trees.

Images