Automatic identification method of artificial afforestation pits based on artificial intelligence remote sensing interpretation technology
Through artificial intelligence remote sensing technology and deep learning models, artificial afforestation pits are automatically identified and accepted, solving the problem of missed inspection and missed inspection by traditional acceptance methods, and achieving efficient and accurate pit management and resource allocation.
Patent Information
- Application Number
- CN202411311255.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-20
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-09-20
AI Technical Summary
The traditional method of artificial afforestation pit acceptance is time-consuming and labor-intensive, and it is prone to missed or missed inspections, and lacks efficient automatic identification methods.
Using a method based on artificial intelligence remote sensing interpretation technology, the image data is collected through drones, and the deep learning model intelligently extracts pit information, combining shape deviation, neighboring cave distance, density deviation and terrain jump index to generate standard scores and sorting coefficients. Automatically divide afforestation and rectification areas and integrate them into the geographical information system for visual processing.
Effectively avoid missed inspections or misinspections, improve the accuracy and efficiency of pit acceptance, simplify management processes, and improve the reasonable allocation and management efficiency of afforestation resources.
Smart Images

Figure CN118865181B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of automatic pit identification, and in particular to an automatic afforestation pit identification method based on artificial intelligence remote sensing interpretation technology. Background Art
[0002] Artificial afforestation projects are important means of ecological restoration and resource protection, and the acceptance of pits is one of the important links in the project. Traditional acceptance methods are time-consuming and labor-intensive, and are prone to missed detections or false detections. Based on this, the present invention proposes an automatic identification method for artificial afforestation pits based on artificial intelligence remote sensing interpretation technology. By combining remote sensing image acquisition, deep learning models, and afforestation area zoning management, the pits in the afforestation area are intelligently identified, and the pits are automatically accepted, which effectively avoids missed detections or false detections and improves the accuracy and efficiency of pit acceptance. Summary of the invention
[0003] The purpose of the present invention is to provide an automatic identification method for artificial afforestation pits based on artificial intelligence remote sensing interpretation technology to overcome the shortcomings of the background technology.
[0004] In order to achieve the above object, the present invention provides the following technical solution: an automatic identification method for artificial afforestation pits based on artificial intelligence remote sensing interpretation technology, the identification method comprising the following steps:
[0005] The recognition system divides the afforestation area into multiple sub-areas. UAVs are used to collect image data for each sub-area. The collected image data are pre-processed and input into the trained deep learning model. The deep learning model intelligently extracts pits from the image data and identifies the pit information in the sub-area images.
[0006] Obtaining the shape deviation and the distance to the neighboring pits of each pit, generating a standard score for each pit according to the shape deviation and the distance to the neighboring pits, and classifying the pits into a non-standard set and a standard set based on the comparison result between the standard score and the score threshold;
[0007] The standard floating value of the pits in the sub-area is calculated according to the standard score, and the standard floating value is combined with the pit density deviation and terrain runout index of the sub-area to generate a sorting coefficient for the sub-area. Based on the comparison result between the sorting coefficient and the coefficient threshold, all sub-areas are divided into an afforestation set and a rectification set.
[0008] After sorting the sub-areas in the afforestation set and rectification set according to the sorting coefficient, the pit information in the afforestation set, rectification set and non-standard set is integrated into the geographic information system, and the geographic information system visualizes the pit information in the afforestation set, rectification set and non-standard set and displays it to the administrator.
[0009] In a preferred embodiment, a standard score is generated for each pit according to the shape deviation and the distance to the neighboring pit, and the pits are classified into a non-standard set and a standard set based on the comparison result between the standard score and the score threshold, including the following steps:
[0010] The shape deviation and the neighboring hole distance are normalized so that the value range of the shape deviation and the neighboring hole distance is mapped to [0.1], and the normalized value of the shape deviation and the normalized value of the neighboring hole distance are obtained. The normalized value of the shape deviation and the normalized value of the neighboring hole distance are summed to obtain the standard score. The standard score is compared with the score threshold. The score threshold is used to determine whether the pit meets the standard. If the standard score is greater than the score threshold, the pit is judged to be substandard and is classified into the non-standard set. If the standard score is less than or equal to the score threshold, the pit is judged to be substandard and is classified into the standard set.
[0011] In a preferred embodiment, the standard floating assignment of pits in the sub-area is calculated based on the standard score, including the following steps:
[0012] After obtaining the standard scores of all pits in the sub-area, a score set is established, the average score and the score discrete value in the score set are calculated, and a standard floating assignment is generated based on the average score and the score discrete value;
[0013] If the average score is less than or equal to the score threshold, and the score discrete value is less than or equal to the discrete threshold, the standard floating assignment ; If the average score is less than or equal to the score threshold, and the score discrete value is greater than the discrete threshold, the standard floating assignment ; If the average score is greater than the score threshold, and the score discrete value is greater than the discrete threshold, the standard floating assignment 2; If the average score is greater than the score threshold, and the score discrete value is less than or equal to the discrete threshold, the standard floating assignment 0.
[0014] In a preferred embodiment, all sub-areas are divided into an afforestation set and a rectification set according to the comparison result between the sorting coefficient and the coefficient threshold, including the following steps:
[0015] The standard floating assignment is combined with the pit density deviation and terrain runout index of the sub-region to generate the sorting coefficient for the sub-region. The calculation expression of the sorting coefficient is: , where is the sorting coefficient, Assign a standard float value, is the pit density deviation, is the terrain runout index, , are the proportional coefficients of the pit density deviation and the terrain runout index, respectively, and , All are greater than 0;
[0016] The obtained ranking coefficient is compared with the preset ranking threshold, which is used to determine whether the sub-area needs afforestation or rectification;
[0017] If the sorting coefficient is greater than or equal to the sorting threshold, it is judged that the sub-region needs afforestation and the sub-region is included in the afforestation set. If the sorting coefficient is less than the sorting threshold, it is judged that the sub-region needs rectification and the sub-region is included in the rectification set.
[0018] In a preferred embodiment, the sub-regions are sorted according to the sorting coefficients in the afforestation set and the rectification set, including the following steps:
[0019] In the afforestation set, all sub-areas are sorted from large to small according to the sorting coefficient, and in the rectification set, all sub-areas are sorted from small to large according to the sorting coefficient.
[0020] In a preferred embodiment, the logic for obtaining the pit density deviation is as follows: obtaining a standard pit density and an actual pit density of a sub-region, subtracting the actual pit density from the standard pit density to obtain a density difference, and taking an absolute value of the density difference as the pit density deviation;
[0021] The calculation logic of the terrain runout index is: divide the sub-area into several grids and obtain the sub-area elevation data, and express it as a matrix , then the calculation expression of terrain runout index is: , where is the number of rows of the sub-area grid, is the number of columns of the sub-area grid, Representing the region Row and The elevation value of the column, represents the average elevation value in the sub-area, and .
[0022] In a preferred embodiment, the average score and the discrete value of the score in the score set are calculated, and the expression is: , where is the average rating, is the discrete value of the score, is the number of pits in the sub-area, For the Standard score for each pit.
[0023] In a preferred embodiment, the calculation expression of the shape deviation is: , where represents the shape deviation, A and B represent the point sets of the standard pit and the actual pit boundary, respectively. Yes and Point The distance between represents the maximum value in the point set, represents the minimum value in the point set, Indicates selecting the maximum value in a set;
[0024] The calculation logic of the neighboring pit distance is as follows: obtain the actual distance between the current pit and all surrounding pits, obtain the standard distance between the pits, subtract the standard distance from the actual distance to obtain the distance difference, take the absolute value of the distance difference, and select the maximum absolute value of the distance difference as the neighboring pit distance.
[0025] In a preferred embodiment, the deep learning model intelligently extracts pits in the image data and identifies the pit information in the sub-region image, including the following steps:
[0026] The image data is preprocessed, including geometric correction, radiation correction, image stitching, denoising and enhancement. The preprocessed image data is input into the trained deep learning model. The deep learning model automatically extracts features in the image and classifies them. The convolution layer is used to extract low-level and high-level features in the image, and the extracted features are analyzed through the fully connected layer to identify the location of pits in the image. The identified pit information is marked on the image, including the location, shape and size information of the pit.
[0027] In a preferred embodiment, the geographic information system visualizes the pit information in the afforestation set, rectification set and non-standard set and displays it to the administrator; an afforestation list is generated based on the afforestation set, and the afforestation order of the sub-area is selected according to the afforestation list; a rectification list is generated based on the rectification set, and the rectification order of the sub-area is selected according to the afforestation list; all pits in the non-standard set are displayed in a highlighted color in the geographic information system, and all pits in the non-standard set are pits that fail acceptance; on the visual map, all pits that fail acceptance are displayed as red dots.
[0028] In the above technical solution, the technical effects and advantages provided by the present invention are:
[0029] 1. The present invention uses a deep learning model to intelligently extract pits from image data, identifies pit information in sub-region images, generates a standard score for each pit based on shape deviation and adjacent pit distance, and classifies pits into a non-standard set and a standard set based on the comparison result between the standard score and the score threshold. The standard floating assignment of pits in the sub-region is calculated based on the standard score, and the standard floating assignment is combined with the pit density deviation and terrain runout index of the sub-region to generate a sorting coefficient for the sub-region. All sub-regions are divided into an afforestation set and a rectification set based on the comparison result between the sorting coefficient and the coefficient threshold. After sorting the sub-regions in the afforestation set and the rectification set according to the sorting coefficient, the pit information in the afforestation set, the rectification set and the non-standard set is integrated into the geographic information system. The recognition system combines remote sensing image acquisition, deep learning models and afforestation area zoning management to intelligently identify pits in the afforestation area, and then automatically inspects the pits, effectively avoiding missed detection or false detection problems, and improving the accuracy and efficiency of pit acceptance.
[0030] 2. The present invention visualizes the pit information in the afforestation set, rectification set and non-standard set through the geographic information system and displays it to the administrator. An afforestation list is generated based on the afforestation set, and the afforestation order of the sub-area is selected according to the afforestation list. A rectification list is generated based on the rectification set, and the rectification order of the sub-area is selected according to the afforestation list. All pits in the non-standard set are displayed in a highlighted color in the geographic information system. All pits in the non-standard set are pits that fail the acceptance. All pits that fail the acceptance are displayed as red dots on the visual map, which facilitates the rational allocation of afforestation management resources, improves management efficiency and further accelerates afforestation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.
[0032] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION
[0033] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0034] Example 1: Please refer to Figure 1 As shown, the method for automatic identification of artificial afforestation pits based on artificial intelligence remote sensing interpretation technology described in this embodiment includes the following steps:
[0035] The recognition system divides the afforestation area into multiple sub-areas. Each sub-area collects image data through drones, pre-processes the collected image data, and inputs the pre-processed image data into the trained deep learning model. The deep learning model intelligently extracts the pits in the image data, identifies the pit information in the sub-area image, obtains the shape deviation and neighboring pit distance of each pit, generates a standard score for each pit based on the shape deviation and neighboring pit distance, and divides the pits into non-standard sets and standard sets based on the comparison results of the standard score and the score threshold. The standard floating assignment of the pits in the sub-area is calculated according to the standard score, and the standard floating assignment is combined with the pit density deviation and terrain runout index of the sub-area to generate a sorting coefficient for the sub-area. According to the comparison results of the sorting coefficient and the coefficient threshold, all sub-areas are sorted. The area is divided into an afforestation set and a rectification set. After sorting the sub-areas in the afforestation set and the rectification set according to the sorting coefficient, the pit information in the afforestation set, the rectification set and the non-standard set is integrated into the geographic information system (GIS). The geographic information system visualizes the pit information in the afforestation set, the rectification set and the non-standard set and displays it to the administrator. For example, an afforestation list is generated based on the afforestation set, and the afforestation order of the sub-area is selected according to the afforestation list. A rectification list is generated based on the rectification set, and the rectification order of the sub-area is selected according to the afforestation list. In the geographic information system, all pits in the non-standard set are displayed in a highlighted color, and all pits in the non-standard set are unqualified pits. For example, on the visual map, all unqualified pits are displayed as red dots for easy management.
[0036] The present application uses a deep learning model to intelligently extract pits from image data, identifies pit information in sub-region images, generates a standard score for each pit based on shape deviation and distance to adjacent pits, divides the pits into a non-standard set and a standard set based on the comparison result between the standard score and the score threshold, calculates the standard floating value of the pits in the sub-region based on the standard score, and combines the standard floating value with the pit density deviation and terrain runout index of the sub-region to generate a sorting coefficient for the sub-region, divides all sub-regions into an afforestation set and a rectification set based on the comparison result between the sorting coefficient and the coefficient threshold, and after sorting the sub-regions in the afforestation set and the rectification set based on the sorting coefficient, integrates the pit information in the afforestation set, the rectification set and the non-standard set into a geographic information system. The recognition system combines remote sensing image acquisition, deep learning models and afforestation area zoning management to intelligently identify pits in the afforestation area, and then automatically inspects the pits, effectively avoiding missed or false detection problems, and improving the accuracy and efficiency of pit acceptance.
[0037] This application visualizes the pit information in the afforestation set, rectification set and non-standard set through the geographic information system and displays it to the administrator. It generates an afforestation list based on the afforestation set, and selects the afforestation order of the sub-area based on the afforestation list. It generates a rectification list based on the rectification set, and selects the rectification order of the sub-area based on the afforestation list. All pits in the non-standard set are displayed in a highlighted color in the geographic information system. All pits in the non-standard set are pits that failed the acceptance. All pits that failed the acceptance are displayed as red dots on the visual map, which facilitates the rational allocation of afforestation management resources, improves management efficiency and further accelerates afforestation efficiency.
[0038] The recognition system includes a sub-region image acquisition module, a pit recognition module, a pit scoring module, a sub-region set division module, and a visualization module;
[0039] Sub-area image acquisition module: the afforestation area is divided into multiple sub-areas, each sub-area collects image data through drones, and the collected image data is pre-processed;
[0040] Pothole identification module: The preprocessed image data is input into the trained deep learning model, which intelligently extracts potholes from the image data and identifies pothole information in the sub-region image.
[0041] Pit scoring module: obtains the shape deviation and neighboring pit distance of each pit, generates a standard score for each pit based on the shape deviation and neighboring pit distance, and classifies the pits into non-standard sets and standard sets based on the comparison results between the standard score and the scoring threshold;
[0042] Sub-area set division module: Calculate the standard floating value of pits in the sub-area according to the standard score, and combine the standard floating value with the pit density deviation and terrain runout index of the sub-area to generate a sorting coefficient for the sub-area. Based on the comparison result between the sorting coefficient and the coefficient threshold, all sub-areas are divided into afforestation set and rectification set;
[0043] Visualization module: After sorting the sub-areas in the afforestation set and rectification set according to the sorting coefficient, the pit information in the afforestation set, rectification set and non-standard set is integrated into the geographic information system (GIS). The geographic information system visualizes the pit information in the afforestation set, rectification set and non-standard set and displays it to the administrator.
[0044] Embodiment 2: The recognition system divides the afforestation area into multiple sub-areas, and each sub-area collects image data through a drone, including the following steps:
[0045] According to the area and terrain characteristics of the afforestation area, the entire area is divided into multiple sub-areas. The purpose of division is to facilitate management and data collection, to ensure that the image coverage of each sub-area is comprehensive and without omissions. Drones are used to perform tasks according to the flight plan to collect high-resolution image data of each sub-area. Drones can be equipped with multi-spectral or hyper-spectral cameras to obtain rich spectral information, providing a data basis for subsequent pit identification.
[0046] The collected image data is preprocessed, and the preprocessed image data is input into the trained deep learning model. The deep learning model intelligently extracts the pits in the image data and identifies the pit information in the sub-region image, including the following steps:
[0047] The image data is preprocessed, including geometric correction, radiation correction, image stitching, denoising and enhancement, and the preprocessed image data is input into the trained deep learning model. The deep learning model is usually based on the architecture of convolutional neural network (CNN), which can automatically extract features in the image and classify them. The deep learning model uses the convolution layer to extract low-level features (such as edges, textures, etc.) and high-level features (such as the shape and shadow of pits) in the image, and analyzes the extracted features through the fully connected layer to identify the location of the pits in the image. According to different features, the model can further classify the type of pits (such as size, shape, etc.), and mark the identified pit information on the image, including detailed information such as the location, shape and size of the pits. The output of the deep learning model can be in the form of annotation boxes, segmentation masks, etc. for further analysis and processing.
[0048] The training of a deep learning model consists of the following steps:
[0049] Collect images of afforestation pits in various environments. The larger the amount of data, the stronger the generalization ability of the model. The image data is annotated, including information such as the location, shape and type of the pits. These annotated data are used to supervise the model to learn features and standardize the images, such as adjusting the image size, normalizing pixel values, and data enhancement (such as rotation, flipping, adjusting brightness, etc.) to increase data diversity;
[0050] Select the VGG convolutional neural network architecture, which includes multiple convolutional layers, pooling layers, and fully connected layers;
[0051] Convolution layer: used to extract image features. It scans the image through the convolution kernel to extract low-level features such as edges and textures.
[0052] Pooling layer: Reduces the size of the feature map, reduces the number of parameters and computation, while retaining important features.
[0053] Fully connected layer: Integrate the extracted features and output classification results or regression values.
[0054] Using non-linear activation functions such as ReLU introduces non-linearity, enabling the model to learn complex features.
[0055] Define the loss function: Use the cross entropy loss function (for classification problems).
[0056] Optimizer selection: The model parameters are adjusted by the stochastic gradient descent (SGD) algorithm to minimize the cross entropy loss function.
[0057] Hyperparameter settings: set learning rate, batch size, and number of training rounds. The learning rate controls the step size of the model weight update, and the batch size determines the amount of data used for each update.
[0058] Forward propagation: Pass the input data forward through the network to obtain the output prediction value.
[0059] Backpropagation: Based on the difference between the predicted value and the true value, the gradient is calculated through the cross entropy loss function, and then the deep learning model parameters are updated through backpropagation.
[0060] During the training process, the validation set is used to evaluate the performance of the model to avoid overfitting. The hyperparameters and architecture of the model are adjusted according to the performance of the validation set, and the final performance of the model is evaluated using an independent test set. The performance on the test set can reflect the generalization ability of the model. According to the results of the validation set, hyperparameters such as learning rate and batch size are adjusted to achieve the best performance. The generalization ability of the model can be improved by increasing the training data or using data augmentation techniques.
[0061] In the above steps, the training of the deep learning model belongs to the prior art and will not be described in detail here. In this application, the specific example scheme is as follows:
[0062] Assume that 10,000 images of afforestation pits are collected, including pits under different lighting conditions, terrains and climate environments. The data annotation includes information about the location (bounding box), shape and type (such as circle, ellipse, etc.). All images are adjusted to a uniform size, such as 224x224 pixels, to match the input requirements of the VGG network, and operations such as rotation, flipping, and brightness adjustment are performed. For example, each image is rotated ±10 degrees, horizontally and vertically flipped, etc., to expand the diversity of the data set.
[0063] VGG Convolutional Neural Network Architecture:
[0064] Convolutional layer: Use a 3x3 convolution kernel, a small step size of 1, and zero padding to keep the feature map size unchanged. For example, after the input image of 224x224 passes through the convolutional layer, the output is still 224x224.
[0065] Pooling layer: Use a 2x2 max pooling layer with a stride of 2 to halve the feature map size. For example, a 224x224 feature map becomes 112x112 after one pooling.
[0066] Fully connected layer: The final fully connected layer is used to output the classification results. Assuming there are 3 categories (pit type A, B, C), the output layer has 3 neurons.
[0067] Use the ReLU activation function: , represents the linear output of the convolutional layer or the fully connected layer, that is, the unactivated output value, Represents the output value after activation, using the ReLU function for activation, which introduces nonlinearity and enables the model to learn more complex features.
[0068] Cross entropy loss function (for classification problems): , is the one-hot encoding of the true label, is the predicted probability of the model.
[0069] The model parameters are adjusted by the stochastic gradient descent (SGD) algorithm: , where are the updated model parameters, are the model parameters before updating, is the learning rate, is the cross entropy loss function About parameters The gradient of represents the partial derivative of the loss function with respect to each parameter.
[0070] Learning rate: Set the learning rate to 0.001.
[0071] Batch size: Set the batch size to 32.
[0072] Number of training rounds: Train for 50 epochs.
[0073] The input data is propagated through the layers of the network to calculate the predicted probability of each class. According to the cross-entropy loss function, the gradient is calculated and the model parameters are updated through backpropagation. After each epoch, the validation set is used to calculate the loss and accuracy, monitor overfitting, and finally an independent test set is used to evaluate the model performance to obtain indicators such as accuracy, precision, and recall.
[0074] Assume that during a training process, the accuracy of the model on the validation set increases from 85% in the 10th epoch to 92% in the 50th epoch, but the accuracy on the test set is only 88%, which indicates that the model may be slightly overfitted on the validation set. To improve the generalization ability, you can:
[0075] Increase the size of the dataset from 10,000 to 30,000 images through data augmentation;
[0076] Use stronger regularization methods, such as Dropout, to randomly ignore some neurons to prevent overfitting.
[0077] Obtaining the shape deviation and the distance to the neighboring pits of each pit, generating a standard score for each pit according to the shape deviation and the distance to the neighboring pits, and classifying the pits into a non-standard set and a standard set based on the comparison result between the standard score and the score threshold, including the following steps:
[0078] The calculation expression of shape deviation is: , where represents the shape deviation, A and B represent the point sets of the standard pit and the actual pit boundary, respectively. Yes and Point The distance between represents the maximum value in the point set, represents the minimum value in the point set, It means selecting the maximum value in the set. A smaller shape deviation means that the two shapes are more similar, while a larger shape deviation means a larger difference. That is, the larger the shape deviation, the more the actual shape of the pit deviates from the standard shape of the pit.
[0079] The calculation logic of the neighboring pit distance is as follows: obtain the actual distance between the current pit and all surrounding pits, obtain the standard distance between pits, subtract the standard distance from the actual distance to obtain the distance difference, take the absolute value of the distance difference, and select the maximum absolute value of the distance difference as the neighboring pit distance. The larger the neighboring pit distance, the closer or farther the current pit is from the surrounding pits, which will affect the subsequent tree planting project.
[0080] The shape deviation and the neighboring hole distance are normalized so that the value range of the shape deviation and the neighboring hole distance is mapped to [0.1], and the normalized value of the shape deviation and the normalized value of the neighboring hole distance are obtained. The normalized value of the shape deviation and the normalized value of the neighboring hole distance are summed to obtain the standard score. The larger the standard score, the less the pit as a whole meets the standard.
[0081] The standard score is compared with the scoring threshold. The scoring threshold is used to determine whether the pit meets the standard. If the standard score is greater than the scoring threshold, the pit is judged to be substandard and is classified into the non-standard set. If the standard score is less than or equal to the scoring threshold, the pit is judged to be substandard and is classified into the standard set.
[0082] The standard floating value assignment for the pits in the sub-area is calculated based on the standard score, including the following steps:
[0083] After obtaining the standard scores of all pits in the sub-area, a score set is established, and the average score and score discrete value in the score set are calculated. The expression is: , where is the average rating, is the discrete value of the score, is the number of pits in the sub-area, For the The larger the score dispersion value, the greater the difference in the standard scores between the pits in the sub-region; the smaller the score dispersion value, the smaller the difference in the standard scores between the pits in the sub-region.
[0084] Generate standard floating values based on average ratings and discrete rating values;
[0085] If the average score is less than or equal to the score threshold, and the discrete value of the score is less than or equal to the discrete threshold, the pits in the analysis sub-area meet the standard as a whole, and the standard floating assignment ;
[0086] If the average score is less than or equal to the score threshold, and the score discrete value is greater than the discrete threshold, the pits in the analysis sub-area meet the standard (but a small number of pits do not meet the standard), and the standard floating assignment ;
[0087] If the average score is greater than the score threshold, and the score discrete value is greater than the discrete threshold, the pits in the analysis sub-area do not meet the standard (but a small number of pits meet the standard), and the standard floating assignment 2;
[0088] If the average score is greater than the score threshold, and the score discrete value is less than or equal to the discrete threshold, the pits in the analysis sub-area do not meet the standard as a whole, and the standard floating assignment 0.
[0089] The standard floating assignment is combined with the pit density deviation and terrain runout index of the sub-region to generate a sorting coefficient for the sub-region. All sub-regions are divided into an afforestation set and a rectification set based on the comparison result between the sorting coefficient and the coefficient threshold, including the following steps:
[0090] The standard floating assignment is combined with the pit density deviation and terrain runout index of the sub-region to generate the sorting coefficient for the sub-region. The calculation expression of the sorting coefficient is: , where is the sorting coefficient, Assign a standard float value, is the pit density deviation, is the terrain runout index, , are the proportional coefficients of the pit density deviation and the terrain runout index, respectively, and , All are greater than 0;
[0091] The larger the ranking coefficient, the more favorable the sub-region is for afforestation (i.e., the smaller the rectification requirement), and the smaller the ranking coefficient, the more rectification is needed in the sub-region. The obtained ranking coefficient is compared with the preset ranking threshold, which is used to determine whether the sub-region needs afforestation or rectification.
[0092] If the sorting coefficient is greater than or equal to the sorting threshold, it is judged that the sub-region needs afforestation and the sub-region is included in the afforestation set. If the sorting coefficient is less than the sorting threshold, it is judged that the sub-region needs rectification and the sub-region is included in the rectification set.
[0093] The logic for obtaining the pit density deviation is as follows: obtain the standard pit density and the actual pit density of the sub-area, subtract the actual pit density from the standard pit density to obtain the density difference, and take the absolute value of the density difference as the pit density deviation. The larger the pit density deviation, the greater or lesser the pit density in the sub-area, and rectification is required.
[0094] The logic for obtaining the actual pit density is: use the measurement tool in the geographic information system (GIS) or obtain the sub-region area from the regional attribute table. In GIS, the sub-region is usually represented in the form of a polygon, and the sub-region area of the polygon can be automatically calculated by the system. This step belongs to the existing technology. In GIS, all pit points in the sub-region can be selected through the spatial query (SpatialQuery) function, and then the number of pits can be counted. The actual pit density is obtained by dividing the actual number of pits by the sub-region area.
[0095] The calculation logic of the terrain runout index is: divide the sub-area into several grids, obtain the sub-area elevation data, and express it as a matrix ,in, Representing the region Row and The calculation expression of terrain runout index is: , where is the number of rows of the sub-area grid, is the number of columns of the sub-area grid, Representing the region Row and The elevation value of the column, represents the average elevation value in the sub-area, and The terrain runout index is a quantitative indicator used to describe the complexity and undulation of the terrain. A higher terrain runout index indicates that the terrain changes dramatically, such as hills, canyons and other significant terrain features, which need to be rectified; a lower terrain runout index indicates that the terrain is relatively flat, such as plains and basins, and no rectification is required.
[0096] After sorting the sub-areas in the afforestation set and the rectification set according to the sorting coefficient, the pit information in the afforestation set, the rectification set and the non-standard set is integrated into the geographic information system (GIS), including the following steps:
[0097] In the afforestation set, all sub-areas are sorted from large to small according to the sorting coefficient, and in the rectification set, all sub-areas are sorted from small to large according to the sorting coefficient.
[0098] Import the pit information in the afforestation set, rectification set and non-standard set into GIS software, such as ArcGIS, QGIS, etc., to create a geographic feature layer, including the location points, boundary polygons, etc. of the pits. In GIS, associate an attribute table with each geographic feature (pit), containing detailed information about the pits (such as sorting coefficient, classification label, terrain information, etc.).
[0099] The geographic information system visualizes the pit information in the afforestation set, rectification set and non-standard set and displays it to the administrator. For example, an afforestation list is generated based on the afforestation set, and the afforestation order of the sub-area is selected based on the afforestation list. A rectification list is generated based on the rectification set, and the rectification order of the sub-area is selected based on the afforestation list. In the geographic information system, all the pits in the non-standard set are displayed in a highlighted color. All the pits in the non-standard set are unqualified pits. For example, on the visualized map, all unqualified pits are displayed as red dots for easy management.
[0100] Extract all pit information from the afforestation set, sort the pits from large to small according to the sorting coefficient, and generate an afforestation list. The list should contain the ID, location, sorting coefficient and other information of each pit;
[0101] Extract all pit information from the rectification set, sort the pits from small to large according to the sorting coefficient, and generate a rectification list. The list should contain the ID, location, sorting coefficient and other information of each pit;
[0102] Based on the ranking coefficients of the pits in the afforestation list, the afforestation order of the sub-areas is selected and recorded in the GIS system for subsequent operations. Based on the ranking coefficients of the pits in the rectification list, the rectification order of the sub-areas is selected and recorded in the GIS system for subsequent operations.
[0103] Create different layers in GIS to display the pit information in the afforestation set, rectification set and non-standard set. Use different symbols or colors to mark the pits. You can use color gradients to indicate the size of the sorting coefficient. Use different symbols or colors to mark the pits. You can use color gradients to indicate the rectification priority. Use prominent colors (such as red) to display all pits that fail the acceptance. Add a legend to the map so that the administrator can understand the meaning of different colors or symbols. Generate maps and reports based on the visualization results to display the pit information, afforestation order and rectification order.
[0104] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.
[0105] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.
[0106] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.
Claims
1. The method for automatic identification of artificial afforestation pits based on artificial intelligence remote sensing interpretation technology is characterized by: The identification method comprises the following steps: The recognition system divides the afforestation area into multiple sub-areas. UAVs are used to collect image data for each sub-area. The collected image data are pre-processed and input into the trained deep learning model. The deep learning model intelligently extracts pits from the image data and identifies the pit information in the sub-area images. The deep learning model intelligently extracts pits from the image data and identifies the pit information in the sub-region image, including the following steps: Preprocess the image data, including geometric correction, radiation correction, image stitching, denoising and enhancement. Input the preprocessed image data into the trained deep learning model. The deep learning model automatically extracts features from the image and classifies them. The convolution layer is used to extract low-level and high-level features from the image, and the extracted features are analyzed through the fully connected layer to identify the location of pits in the image. The identified pit information is marked on the image, including the location, shape and size of the pit; Obtaining the shape deviation and the distance to the neighboring pits of each pit, generating a standard score for each pit according to the shape deviation and the distance to the neighboring pits, and classifying the pits into a non-standard set and a standard set based on the comparison result between the standard score and the score threshold; The standard floating value of the pits in the sub-area is calculated according to the standard score, and the standard floating value is combined with the pit density deviation and terrain runout index of the sub-area to generate a sorting coefficient for the sub-area. Based on the comparison result between the sorting coefficient and the coefficient threshold, all sub-areas are divided into an afforestation set and a rectification set. After sorting the sub-areas in the afforestation set and rectification set according to the sorting coefficient, the pit information in the afforestation set, rectification set and non-standard set is integrated into the geographic information system, and the geographic information system visualizes the pit information in the afforestation set, rectification set and non-standard set and displays it to the administrator; The logic for obtaining the pit density deviation is as follows: obtaining the standard pit density and the actual pit density of the sub-region, subtracting the actual pit density from the standard pit density to obtain the density difference, and taking the absolute value of the density difference as the pit density deviation; The calculation logic of the terrain runout index is: divide the sub-area into several grids and obtain the sub-area elevation data, and express it as a matrix , then the calculation expression of terrain runout index is: , where is the number of rows of the sub-area grid, is the number of columns of the sub-area grid, Representing the region Row and The elevation value of the column, represents the average elevation value in the sub-area, and .
2. The method for automatic identification of artificial afforestation pits based on artificial intelligence remote sensing interpretation technology according to claim 1 is characterized in that: Generate a standard score for each pit according to the shape deviation and the distance to the neighboring pit, and classify the pits into a non-standard set and a standard set based on the comparison result between the standard score and the score threshold, including the following steps: The shape deviation and the neighboring hole distance are normalized so that the value range of the shape deviation and the neighboring hole distance is mapped to [0.1], and the normalized value of the shape deviation and the normalized value of the neighboring hole distance are obtained. The normalized value of the shape deviation and the normalized value of the neighboring hole distance are summed to obtain the standard score. The standard score is compared with the score threshold. The score threshold is used to determine whether the pit meets the standard. If the standard score is greater than the score threshold, the pit is judged to be substandard and is classified into the non-standard set. If the standard score is less than or equal to the score threshold, the pit is judged to be substandard and is classified into the standard set.
3. The method for automatic identification of artificial afforestation pits based on artificial intelligence remote sensing interpretation technology according to claim 2 is characterized in that: The standard floating value assignment for the pits in the sub-area is calculated based on the standard score, including the following steps: After obtaining the standard scores of all pits in the sub-area, a score set is established, the average score and the score discrete value in the score set are calculated, and a standard floating assignment is generated based on the average score and the score discrete value; If the average score is less than or equal to the score threshold, and the score discrete value is less than or equal to the discrete threshold, the standard floating assignment ; If the average score is less than or equal to the score threshold, and the score discrete value is greater than the discrete threshold, the standard floating assignment ; If the average score is greater than the score threshold, and the score discrete value is greater than the discrete threshold, the standard floating assignment 2; If the average score is greater than the score threshold, and the score discrete value is less than or equal to the discrete threshold, the standard floating assignment 0.
4. The method for automatic identification of artificial afforestation pits based on artificial intelligence remote sensing interpretation technology according to claim 3 is characterized in that: According to the comparison results between the sorting coefficient and the coefficient threshold, all sub-areas are divided into afforestation sets and rectification sets, including the following steps: The standard floating assignment is combined with the pit density deviation and terrain runout index of the sub-region to generate the sorting coefficient for the sub-region. The calculation expression of the sorting coefficient is: , where is the sorting coefficient, Assign a standard float value, is the pit density deviation, is the terrain runout index, , are the proportional coefficients of the pit density deviation and the terrain runout index, respectively, and , All are greater than 0; The obtained ranking coefficient is compared with the preset ranking threshold, which is used to determine whether the sub-area needs afforestation or rectification; If the sorting coefficient is greater than or equal to the sorting threshold, it is judged that the sub-region needs afforestation and the sub-region is included in the afforestation set. If the sorting coefficient is less than the sorting threshold, it is judged that the sub-region needs rectification and the sub-region is included in the rectification set.
5. The method for automatic identification of artificial afforestation pits based on artificial intelligence remote sensing interpretation technology according to claim 4 is characterized in that: Sorting the sub-areas in the afforestation set and the rectification set according to the sorting coefficient includes the following steps: In the afforestation set, all sub-areas are sorted from large to small according to the sorting coefficient, and in the rectification set, all sub-areas are sorted from small to large according to the sorting coefficient.
6. The method for automatic identification of artificial afforestation pits based on artificial intelligence remote sensing interpretation technology according to claim 5 is characterized in that: Calculate the average score and discrete value of the score in the score set. The expression is: , where is the average rating, is the discrete value of the score, is the number of pits in the sub-area, For the Standard score for each pit.
7. The method for automatic identification of artificial afforestation pits based on artificial intelligence remote sensing interpretation technology according to claim 2 is characterized in that: The calculation expression of the shape deviation is: , where represents the shape deviation, A and B represent the point sets of the standard pit and the actual pit boundary, respectively. Yes and Point The distance between represents the maximum value in the point set, represents the minimum value in the point set, Indicates selecting the maximum value in a set; The calculation logic of the neighboring pit distance is as follows: obtain the actual distance between the current pit and all surrounding pits, obtain the standard distance between the pits, subtract the standard distance from the actual distance to obtain the distance difference, take the absolute value of the distance difference, and select the maximum absolute value of the distance difference as the neighboring pit distance.
8. The method for automatic identification of artificial afforestation pits based on artificial intelligence remote sensing interpretation technology according to claim 7 is characterized in that: The geographic information system visualizes the pit information in the afforestation set, rectification set and non-standard set and displays it to the administrator. It generates an afforestation list based on the afforestation set, and selects the afforestation order of the sub-area based on the afforestation list. It generates a rectification list based on the rectification set, and selects the rectification order of the sub-area based on the afforestation list. All pits in the non-standard set are displayed in a highlighted color in the geographic information system. All pits in the non-standard set are pits that fail the acceptance. All pits that fail the acceptance are displayed as red dots on the visualized map.
