Desertification grassland mousehole identification deep learning method based on neural network model

Abstract

The invention discloses a deep learning method for desertification grassland mouse hole recognition based on a neural network model, and belongs to the field of image recognition. The method solves the problems of low efficiency and poor accuracy of an existing mouse hole recognition method, constructs a mouse hole distribution image set through a high-resolution image obtained by an unmanned aerial vehicle, identifies a desertification grassland mouse hole sample according to the mouse hole distribution image set, can obtain a mouse hole feature data set, and achieves the recognition of a mouse hole based on a neural network model. The mouse hole feature data set is used for training the neural network model, mouse holes can be trained and recognized, then the test set is used for testing the trained neural network model, the performance of the neural network model can be evaluated, the steps are repeated, machine deep learning is carried out, and the accuracy of machine learning is improved. The neural network model parameters can be configured, the accuracy of the neural network model can be verified, the grassland mouse hole recognition deep learning method based on the neural network model is finally obtained, and the recognition efficiency and precision of mouse hole recognition can be improved based on the method.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a deep learning method for identifying rat holes in desertified grasslands based on a neural network model. Background Art

[0002] Traditional methods for identifying grassland rat holes usually rely on manual or semi-automatic visual detection, which not only has low recognition efficiency but also makes it difficult to ensure accuracy and consistency. Therefore, it does not meet existing needs. In this regard, we propose a deep learning method for identifying desertified grassland rat holes based on a neural network model. Summary of the invention

[0003] The purpose of the present invention is to provide a deep learning method for identifying rat holes in desertified grasslands based on a neural network model. The rat hole samples in desertified grasslands are identified by constructing a rat hole distribution image set to obtain a rat hole feature data set. Based on the neural network model, the rat hole feature data set is used to train the neural network model, so that rat holes can be trained and identified. The trained neural network model is then tested using a test set to evaluate the performance of the neural network model. The above steps are repeated to perform machine deep learning, thereby configuring the neural network model parameters and verifying the accuracy of the neural network model. Finally, a deep learning method for identifying grassland rat holes based on a neural network model is obtained, which solves the problems raised in the above background technology.

[0004] To achieve the above object, the present invention provides the following technical solution: a deep learning method for identifying rat holes in desertified grasslands based on a neural network model, comprising the following steps:

[0005] Step 1: Determine the grassland rodent monitoring area and use the drone to fly to the grassland rodent monitoring area;

[0006] Step 2: Use drones to obtain high-resolution images of the grassland rodent monitoring area, pre-process the obtained high-resolution images and construct a rat hole distribution image set, where the high-resolution images are high-spectral resolution images;

[0007] Step 3: Identify the desertified grassland rat hole samples based on the rat hole distribution image set to obtain a rat hole feature dataset, which is used to train the neural network model;

[0008] Step 4: Based on the neural network model, the rathole feature data set is used to train the neural network model, and then the parameters are adjusted to train and identify ratholes. The trained neural network model is then tested using the test set to evaluate the performance of the neural network model.

[0009] Step 5: Repeat steps 3 and 4 to perform machine deep learning, configure neural network model parameters, and verify the accuracy of the neural network model to obtain a deep learning method for identifying prairie mouse holes based on a neural network model.

[0010] Furthermore, the drone carries a hyperspectral imager, which uses the hyperspectral imager to obtain high-resolution images of the grassland rodent pest monitoring area, pre-processes the obtained high-resolution images, and then constructs the pre-processed high-resolution images into a rat hole distribution image set.

[0011] Furthermore, the preprocessing includes:

[0012] Orthophoto stitching: stitch the acquired high-resolution images to obtain orthophotos for subsequent processing;

[0013] Color image segmentation processing: color image segmentation is performed on the spliced ​​orthophotos to separate different parts of the high-resolution images, where the different parts are grasslands and mouse holes;

[0014] Grayscale processing: Convert color high-resolution images into grayscale images to simplify the high-resolution image processing process, while reducing the amount of calculation and increasing the processing speed;

[0015] Image smoothing: Smoothing high-resolution images to reduce noise and details in them and make them smoother for subsequent edge detection.

[0016] Edge detection processing: Use the Sobel operator to extract the edge of the high-resolution image to detect the edge information in the high-resolution image for subsequent rathole recognition and identification.

[0017] Furthermore, the preprocessing further includes:

[0018] Extracting a grayscale image, and traversing each pixel block in the grayscale image to obtain a grayscale value corresponding to each pixel block;

[0019] Obtaining a central grayscale value corresponding to the grayscale image according to the grayscale value of a pixel block in a central area of ​​the grayscale image;

[0020] Dividing the grayscale image according to a preset region size to obtain a plurality of image blocks; wherein the region size is n×n, and the value range of n is 3-9;

[0021] Extract the central gray value corresponding to each image block;

[0022] Obtaining a grayscale adjustment coefficient corresponding to each image block using the central grayscale value corresponding to each image block and the central grayscale value corresponding to the grayscale image;

[0023] The grayscale adjustment coefficient is obtained by the following formula:

[0024]

[0025] Where S represents the grayscale adjustment coefficient; m represents the number of image blocks; H zi Represents the central grayscale value corresponding to the i-th image block; H z represents the central grayscale value corresponding to the grayscale image; x represents the adjustment coefficient, and the adjustment coefficient is obtained by the following formula:

[0026]

[0027] Where x represents the adjustment coefficient; m represents the number of image blocks; H zi Represents the central grayscale value corresponding to the i-th image block; H z Indicates the central gray value corresponding to the grayscale image; H zmax and H zmin Represents the maximum and minimum values ​​of the central grayscale values ​​corresponding to the m image blocks;

[0028] The grayscale value of the pixel block included in each image block is adjusted using the grayscale adjustment coefficient corresponding to each image block.

[0029] Furthermore, the grayscale value of the pixel block included in each image block is adjusted by using the grayscale adjustment coefficient corresponding to each image block, including:

[0030] Extracting the grayscale adjustment coefficient corresponding to each image block;

[0031] Extracting the central grayscale value corresponding to each image block and the central grayscale value corresponding to the grayscale image;

[0032] Obtaining a grayscale adjustment coefficient corresponding to the pixel blocks contained in each image block by using the grayscale value of each pixel block in each image block in combination with the central grayscale value of the corresponding image block and the central grayscale value corresponding to the grayscale image;

[0033] The grayscale adjustment coefficient corresponding to the pixel block contained in each image block is obtained by the following formula:

[0034]

[0035] Wherein, F represents the grayscale adjustment coefficient corresponding to the pixel block contained in each image block; H represents the grayscale value corresponding to the pixel block contained in each image block; H zx Indicates the central gray value of the image block corresponding to the pixel block; H zIndicates the central gray value corresponding to the grayscale image; H zp Represents the average central grayscale value of all image blocks adjacent to the image block where the pixel block is located;

[0036] The grayscale adjustment coefficient corresponding to each image block is combined with the grayscale adjustment coefficient corresponding to the pixel block contained in each image block to adjust the grayscale of the pixel block contained in each image block, and the grayscale value of the adjusted pixel block is obtained by the following formula:

[0037]

[0038] Among them, H t represents the gray value of the adjusted pixel block; F represents the gray adjustment coefficient corresponding to the pixel block contained in each image block; H represents the gray value corresponding to the pixel block contained in each image block; S represents the gray adjustment coefficient; ε represents the preset minimum constant to prevent the numerator from being 0.

[0039] Furthermore, the identification of desertified grassland rat hole samples specifically includes the following steps:

[0040] Based on high-resolution images, spectral information of rat holes and their surroundings is obtained;

[0041] Based on the spectral characteristics of rat holes and other grassland features in their surroundings, a rat hole index is proposed.

[0042] According to the proposed rat hole index, the rat hole distribution image set is classified and identified to determine the location and number of rat holes, so as to identify the rat hole samples in desertified grasslands.

[0043] By identifying rat hole samples in desertified grasslands, a rat hole characteristic dataset was obtained.

[0044] Furthermore, the rathole identification results are verified by comparing them with the real ground data and repeating the identification method to ensure the accuracy of the rathole identification results. After verification, the rathole feature data set is obtained and applied to the training of the neural network model.

[0045] Furthermore, the step 4 specifically includes the following steps:

[0046] Model construction: Convolutional neural network is used as the basic model, and the network structure is designed according to the requirements of rathole identification tasks;

[0047] Training model: Based on the neural network model, the rathole feature data set is used to train the neural network model, and then the neural network model is verified using the validation set prepared in advance to adjust the neural network model parameters and improve the performance of the neural network model;

[0048] Model testing: Use the test set to test the trained neural network model to evaluate the accuracy of the neural network model.

[0049] Furthermore, the training model comprises the following steps:

[0050] Forward propagation: Forward propagation is performed during the training process. The characteristic variables of the input training samples are processed through each layer of the neural network model to finally obtain the output result.

[0051] Calculate loss: Compare the output of the neural network model with the true label value and calculate the difference between the two;

[0052] Back propagation: Perform the back propagation process and calculate the gradient of the loss function to the neural network model parameters based on the calculated loss through optimization algorithms such as gradient descent;

[0053] Parameter update: Based on the calculated gradient information, the weights and bias parameters of the neural network model are updated to ensure that the parameter adjustment is not too large or too small;

[0054] Iterative optimization: The above steps will be iterated multiple times. Through multiple iterations, the neural network model gradually learns the mapping relationship from input to output.

[0055] Furthermore, the test set used for the model test is prepared in advance, and the data in the test set does not overlap with the data in the rathole feature data set. Before conducting the model test, the test set is normalized to meet the input requirements of the neural network model, and the processed test set is input into the neural network model to obtain the test results. According to the analysis results, the accuracy, precision and recall rate indicators of the neural network model are calculated to evaluate the performance of the neural network model.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] The present invention uses a drone to fly to a grassland rodent monitoring area to obtain high-resolution images, and uses this to construct a rat hole distribution image set. According to the rat hole distribution image set, desertified grassland rat hole samples are identified, and a rat hole feature data set can be obtained. Based on a neural network model, the neural network model is trained using the rat hole feature data set, so that rat holes can be trained and identified. The trained neural network model is then tested using a test set, so that the performance of the neural network model can be evaluated. The above steps are repeated to perform machine deep learning, configure neural network model parameters and verify the accuracy of the neural network model, and finally a grassland rat hole identification deep learning method based on the neural network model is obtained. Based on this method, the recognition efficiency and accuracy of rat holes can be greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 The present invention is a flow chart of a deep learning method for identifying rat holes in desertified grasslands based on a neural network model. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0060] In order to solve the technical problem that the existing prairie mouse hole identification methods usually rely on manual or semi-automatic visual detection, which is not only inefficient but also difficult to ensure accuracy and consistency, please refer to Figure 1 , this embodiment provides the following technical solutions:

[0061] The deep learning method for identifying rat holes in desertified grasslands based on a neural network model includes the following steps:

[0062] Step 1: Determine the grassland rodent monitoring area and use the drone to fly to the grassland rodent monitoring area;

[0063] Step 2: Use drones to obtain high-resolution images of the grassland rodent monitoring area, pre-process the obtained high-resolution images and construct a rat hole distribution image set, where the high-resolution images are high-spectral resolution images;

[0064] Step 3: Identify the desertified grassland rat hole samples based on the rat hole distribution image set to obtain a rat hole feature dataset, which is used to train the neural network model;

[0065] Step 4: Based on the neural network model, the rathole feature data set is used to train the neural network model, and then the parameters are adjusted to train and identify ratholes. The trained neural network model is then tested using the test set to evaluate the performance of the neural network model.

[0066] Step 5: Repeat steps 3 and 4 to perform machine deep learning, configure neural network model parameters, and verify the accuracy of the neural network model to obtain a deep learning method for identifying prairie mouse holes based on a neural network model.

[0067] The drone carries a hyperspectral imager, which is used to obtain high-resolution images of the grassland rodent monitoring area, pre-process the obtained high-resolution images, and then construct the pre-processed high-resolution images into a rat hole distribution image set.

[0068] Among them, preprocessing includes:

[0069] Orthophoto stitching: stitching the acquired high-resolution images to obtain orthophotos. The purpose of this process is to merge the images into a complete image for subsequent processing.

[0070] Color image segmentation processing: color image segmentation is performed on the spliced ​​orthophotos to separate different parts of the high-resolution images, where the different parts are grasslands and mouse holes;

[0071] Grayscale processing: Convert color high-resolution images into grayscale images to simplify the high-resolution image processing process, while reducing the amount of calculation and increasing the processing speed;

[0072] Image smoothing: Smoothing high-resolution images to reduce noise and details in them and make them smoother for subsequent edge detection.

[0073] Edge detection processing: Use the Sobel operator to extract the edge of the high-resolution image to detect the edge information in the high-resolution image for subsequent rathole recognition and identification.

[0074] The technical effect of the above content is: using a drone carrying a hyperspectral imager to collect hyperspectral data of surface rat hole samples in desertified grasslands, so as to obtain high-resolution images. This step is the basis of the entire identification process. Through the hyperspectral imaging technology of the drone, the target object can be continuously imaged in the spectral range from ultraviolet to near-infrared, and the spectral information of the rat hole and its surrounding environment can be obtained at the same time. This technology improves the efficiency and accuracy of data collection. By analyzing the spectral information of the rat hole and its surrounding environment, a rat hole index (RHI) can be proposed. This index can effectively improve the subsequent recognition accuracy of grassland rat holes. Then, the obtained high-resolution images are preprocessed by orthophoto stitching, color image segmentation, grayscale processing, image smoothing and edge detection. Through these preprocessing steps, the accuracy and efficiency of subsequent rat hole identification can be greatly improved. These preprocessing steps not only involve basic image processing techniques, but also include optimization processing for specific application scenarios (such as grassland rat hole monitoring) to ensure the accuracy and reliability of the final results.

[0075] Specifically, the preprocessing further includes:

[0076] Extracting a grayscale image, and traversing each pixel block in the grayscale image to obtain a grayscale value corresponding to each pixel block;

[0077] According to the grayscale value of the pixel block in the central area of ​​the grayscale image, obtaining the central grayscale value corresponding to the grayscale image;

[0078] Dividing the grayscale image according to a preset region size to obtain a plurality of image blocks; wherein the region size is n×n, and the value range of n is 3-9;

[0079] Extract the central gray value corresponding to each image block;

[0080] Obtaining a grayscale adjustment coefficient corresponding to each image block using the central grayscale value corresponding to each image block and the central grayscale value corresponding to the grayscale image;

[0081] The grayscale adjustment coefficient is obtained by the following formula:

[0082]

[0083] Where S represents the grayscale adjustment coefficient; m represents the number of image blocks; H zi Represents the central grayscale value corresponding to the i-th image block; H z represents the central grayscale value corresponding to the grayscale image; x represents the adjustment coefficient, and the adjustment coefficient is obtained by the following formula:

[0084]

[0085] Where x represents the adjustment coefficient; m represents the number of image blocks; H zi Represents the central grayscale value corresponding to the i-th image block; H z Indicates the central gray value corresponding to the grayscale image; H zmax and H zmin Represents the maximum and minimum values ​​of the central grayscale values ​​corresponding to the m image blocks;

[0086] The grayscale value of the pixel block included in each image block is adjusted using the grayscale adjustment coefficient corresponding to each image block.

[0087] The technical effect of the above technical solution is: by extracting the grayscale image and calculating the grayscale value of each pixel block, and further calculating the central grayscale value of the grayscale image, the above technical solution provides a benchmark for subsequent grayscale adjustment. By dividing the image into multiple n×n image blocks (n ranges from 3 to 9) and calculating the central grayscale value of each image block, the grayscale distribution of the image can be analyzed more carefully. At the same time, the grayscale adjustment coefficient is calculated using the relationship between the central grayscale value of each image block and the central grayscale value of the entire grayscale image. This step aims to perform adaptive adjustments based on the local grayscale features of the image. The calculation formula of the grayscale adjustment coefficient takes into account the difference between the central grayscale value of the image block and the overall central grayscale value, as well as the maximum and minimum values ​​of the central grayscale values ​​of all image blocks, thereby achieving relative balancing of the grayscale values.

[0088] According to the calculated grayscale adjustment coefficient, the grayscale value of the pixel block in each image block is adjusted. This step is intended to make the grayscale distribution of the image more uniform and reduce the situation of local over-brightness or over-darkness. Through this adjustment, the contrast of the image may be improved, making the details in the image clearer. After the above processing, the grayscale distribution of the image will be more uniform, and the contrast will be improved, thereby improving the overall quality of the image. The above technical scheme is beneficial for subsequent image processing tasks, such as feature extraction, target detection, image recognition, etc. At the same time, the n×n image block division (n ranges from 3 to 9) in the above technical scheme provides a flexible processing method, which can be adjusted according to different image sizes and resolutions. The calculation formula of the grayscale adjustment coefficient also takes into account a variety of factors, so that the above technical scheme can adapt to different types of grayscale images.

[0089] In summary, this technical solution achieves optimized processing of grayscale images through a series of steps, improves the quality and contrast of images, and provides a better foundation for subsequent image processing tasks.

[0090] Specifically, adjusting the grayscale value of the pixel block included in each image block by using the grayscale adjustment coefficient corresponding to each image block includes:

[0091] Extracting the grayscale adjustment coefficient corresponding to each image block;

[0092] Extracting the central grayscale value corresponding to each image block and the central grayscale value corresponding to the grayscale image;

[0093] Obtaining a grayscale adjustment coefficient corresponding to the pixel blocks contained in each image block by using the grayscale value of each pixel block in each image block in combination with the central grayscale value of the corresponding image block and the central grayscale value corresponding to the grayscale image;

[0094] The grayscale adjustment coefficient corresponding to the pixel block contained in each image block is obtained by the following formula:

[0095]

[0096] Wherein, F represents the grayscale adjustment coefficient corresponding to the pixel block contained in each image block; H represents the grayscale value corresponding to the pixel block contained in each image block; H zx Indicates the central gray value of the image block corresponding to the pixel block; H z Indicates the central gray value corresponding to the grayscale image; H zp Represents the average central grayscale value of all image blocks adjacent to the image block where the pixel block is located;

[0097] The grayscale adjustment coefficient corresponding to each image block is combined with the grayscale adjustment coefficient corresponding to the pixel block contained in each image block to adjust the grayscale of the pixel block contained in each image block, and the grayscale value of the adjusted pixel block is obtained by the following formula:

[0098]

[0099] Among them, H t represents the gray value of the adjusted pixel block; F represents the gray adjustment coefficient corresponding to the pixel block contained in each image block; H represents the gray value corresponding to the pixel block contained in each image block; S represents the gray adjustment coefficient; ε represents the preset minimum constant to prevent the numerator from being 0.

[0100] The technical effect of the above technical solution is: by calculating the grayscale adjustment coefficient for each pixel block in each image block, the above technical solution achieves a more refined grayscale adjustment. This helps to better retain the detailed information of the image while maintaining the overall grayscale distribution of the image. When calculating the grayscale adjustment coefficient of each pixel block, the average value of the central grayscale value of adjacent image blocks is introduced. This step takes into account the mutual influence between image blocks and helps to maintain the continuity and consistency of the image when adjusting the grayscale.

[0101] The above technical solution combines the grayscale adjustment coefficient at the image block level and the grayscale adjustment coefficient at the pixel block level. By comprehensively applying these two coefficients, a more accurate adjustment of the image grayscale is achieved. This helps to improve the contrast of the image and make the image clearer. When calculating the grayscale value of the adjusted pixel block, a preset minimum constant ε is introduced to prevent the denominator from being zero. This step ensures the stability and reliability of the calculation and avoids erroneous results caused by the denominator being zero. After the above processing, the grayscale distribution of the image will be more uniform, the contrast will be improved, and the detail information will be retained. This helps to improve the overall quality and visual effect of the image, making the image more natural and clear. The formulas and parameters in the above technical solution can be adjusted according to different image characteristics and application requirements. For example, the size of the n×n image block, the preset minimum constant ε and other parameters can be adjusted to adapt to different types of grayscale images and processing tasks.

[0102] In summary, this technical solution achieves more accurate and effective adjustment of grayscale images through refined grayscale adjustment, consideration of the influence of adjacent image blocks, comprehensive application of grayscale adjustment coefficients, and prevention of zero denominator processing. This helps to improve the quality and visual effects of images and provides a better foundation for subsequent image processing tasks.

[0103] The identification of rat hole samples in desertified grasslands includes the following steps:

[0104] Based on high-resolution images, spectral information of rat holes and their surroundings is obtained;

[0105] Based on the spectral characteristics of rat holes and other grassland features in their surroundings, a rat hole index is proposed.

[0106] According to the proposed rat hole index, the rat hole distribution image set is classified and identified to determine the location and number of rat holes, so as to identify the rat hole samples in desertified grasslands.

[0107] By identifying rat hole samples in desertified grasslands, a rat hole characteristic dataset was obtained.

[0108] The technical effect of the above content is: the rat hole index (RHI) is a method for collecting data on desertified grasslands using a drone carrying a hyperspectral imager, which is used to identify rat holes on grasslands. This method improves the recognition accuracy and efficiency of grassland rat holes by proposing RHI. First, based on the high-resolution images acquired and processed by the drone, the spectral information of the rat holes and their surroundings can be obtained. Based on the spectral characteristics of the rat holes and other grassland objects in their surroundings, the rat hole index (RHI) can be proposed. According to the rat hole index (RHI), high-precision recognition of rat holes in desertified grasslands can be achieved. Through the classification and recognition of data by the rat hole index, the location and number of rat holes can be determined, thereby realizing the identification of rat hole samples in desertified grasslands, and finally obtaining a rat hole feature data set. The RHI method can effectively improve the recognition accuracy of grassland rat holes and provide a scientific basis for grassland management and protection. This method not only improves the recognition accuracy, but also greatly improves the work efficiency, which is of great significance for the management and protection of desertified grasslands.

[0109] The rathole identification results are verified by comparing them with the real ground data and repeating the identification method to ensure the accuracy of the rathole identification results. After verification, the rathole feature data set is obtained and used to train the neural network model.

[0110] The technical effect of the above content is: after the rat hole samples in the desertified grassland are identified, it is necessary to verify the rat hole identification results. The verification is carried out by comparing with the real data on the ground and repeating the identification. The accuracy of the rat hole identification results can be tested through verification, thereby ensuring the accuracy of the rat hole identification results. After the rat hole identification results are verified by the above method, the rat hole feature data set can be obtained, and the rat hole feature data set can be used for the subsequent training of the neural network model, so as to improve the accuracy and reliability of the neural network model.

[0111] Step 4 specifically includes the following steps:

[0112] Model construction: Convolutional neural network is used as the basic model, and the network structure is designed according to the requirements of rathole identification tasks;

[0113] Training model: Based on the neural network model, the rathole feature data set is used to train the neural network model, and then the neural network model is verified using the validation set prepared in advance to adjust the neural network model parameters and improve the performance of the neural network model;

[0114] Among them, through the process of training the model, the neural network model can adjust its internal parameters according to the training data to improve the prediction or classification ability of unknown data, which specifically includes the following steps:

[0115] Forward propagation: Forward propagation is performed during the training process. The characteristic variables of the input training samples are processed through each layer of the neural network model to finally obtain the output result. The calculations involved in this process include weighted summation and nonlinear transformation through activation functions.

[0116] Calculate loss: Compare the output of the neural network model with the actual label value and calculate the difference between the two. This difference is usually quantified by the loss function, which measures the gap between the output of the neural network model and the actual result and is the objective function of the optimization process.

[0117] Back propagation: Perform the back propagation process. According to the calculated loss, the gradient of the loss function to the parameters of the neural network model is calculated through optimization algorithms such as gradient descent. These gradient information starts from the output layer and propagates back to the input layer to guide the update of the parameters of the neural network model to reduce the loss function value.

[0118] Parameter update: According to the calculated gradient information, the weight and bias parameters of the neural network model are updated. The parameter update rule usually follows a certain learning rate to ensure that the parameter adjustment is not too large or too small, so as to maintain the stability of the neural network model learning;

[0119] Iterative optimization: The above steps will be repeated multiple times until certain stopping conditions are met, such as reaching a preset number of iterations, the loss value is less than a certain threshold, or the performance of the neural network model on the validation set is no longer significantly improved. Through multiple iterations, the neural network model gradually learns the mapping relationship from input to output;

[0120] Model testing: Use the test set to test the trained neural network model to evaluate the accuracy of the neural network model;

[0121] Among them, the test set used for model testing is prepared in advance, and the data in the test set does not overlap with the rathole feature data set. Before model testing, the test set is normalized to meet the input requirements of the neural network model. The processed test set is input into the neural network model to obtain the test results. According to the analysis results, the accuracy, precision and recall rate indicators of the neural network model are calculated to evaluate the performance of the neural network model.

[0122] The technical effect of the above content is: by adopting a convolutional neural network as a basic model and using a rat hole feature data set to train the neural network model, rat holes can be trained and identified, and then the neural network model is verified using a pre-prepared verification set, so that the neural network model parameters can be adjusted and the performance of the neural network model can be improved, thereby obtaining the best neural network model. Before the actual use of this neural network model, it is necessary to use a test set for testing. The test set is used to evaluate the final performance of the neural network model to ensure that the neural network model can correctly process unseen data. The use of the test set ensures the generalization ability of the neural network model, that is, the neural network model can make accurate predictions when facing new data. The test set used for model testing is prepared in advance, and the data in the test set has not been used in the training process, thereby ensuring the objectivity and accuracy of the test results. The selection of the test set should be random and representative to comprehensively evaluate the performance of the neural network model. By reasonably using the training set, verification set and test set, the performance of the neural network model can be effectively optimized to ensure the generalization ability of the neural network model, thereby improving the prediction or classification accuracy of the neural network model.

[0123] Working principle: By flying drones to the grassland rodent monitoring area, high-resolution images can be obtained, and a rat hole distribution image set can be constructed based on this. The desertified grassland rat hole samples can be identified based on the rat hole distribution image set. Based on the spectral characteristics of the rat holes and other grassland objects in their surroundings, a rat hole index can be proposed. According to the rat hole index, high-precision identification of desertified grassland rat holes can be achieved. After the desertified grassland rat hole samples are identified, the rat hole identification results are verified by comparing with the real data on the ground and repeating the identification method, thereby ensuring the accuracy of the rat hole identification results. Finally, the rat hole feature data set can be obtained. Based on the neural network model, the rat hole feature data set is used to train the neural network model, which can train rat holes. The neural network model is then verified using the validation set prepared in advance, so that the parameters of the neural network model can be adjusted and the performance of the neural network model can be improved. Finally, the neural network model is tested using the test set to evaluate the performance of the neural network model. By using the training set, validation set, and test set reasonably, the performance of the neural network model can be effectively optimized, the generalization ability of the neural network model can be ensured, and the prediction or classification accuracy of the neural network model can be improved. By repeating the above steps, machine deep learning can be performed, the parameters of the neural network model can be configured, and the accuracy of the neural network model can be verified. Finally, a deep learning method for identifying prairie mouse holes based on the neural network model is obtained, and this method can greatly improve the efficiency and accuracy of identifying mouse holes.

[0124] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0125] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

Claims

1. A deep learning method for identifying rat holes in desertified grasslands based on a neural network model, characterized in that: The following steps are involved: Step 1: Determine the grassland rodent monitoring area and use the drone to fly to the grassland rodent monitoring area; Step 2: Use drones to obtain high-resolution images of the grassland rodent monitoring area, pre-process the obtained high-resolution images and construct a rat hole distribution image set, where the high-resolution images are high-spectral resolution images; Step 3: Identify the desertified grassland rat hole samples based on the rat hole distribution image set to obtain a rat hole feature data set, which is used to train the neural network model; Step 4: Based on the neural network model, the rathole feature data set is used to train the neural network model, and then the parameters are adjusted to train and identify ratholes. The trained neural network model is then tested using the test set to evaluate the performance of the neural network model. Step 5: Repeat steps 3 and 4 to perform machine deep learning, configure neural network model parameters, and verify the accuracy of the neural network model to obtain a deep learning method for identifying prairie mouse holes based on a neural network model.

2. The deep learning method for identifying rat holes in desertified grasslands based on a neural network model according to claim 1 is characterized in that: The drone carries a hyperspectral imager, which uses the hyperspectral imager to obtain high-resolution images of the grassland rodent monitoring area, pre-processes the obtained high-resolution images, and then constructs the pre-processed high-resolution images into a rat hole distribution image set.

3. The deep learning method for identifying rat holes in desertified grasslands based on a neural network model according to claim 2 is characterized in that: The preprocessing comprises: Orthophoto stitching: stitch the acquired high-resolution images to obtain orthophotos for subsequent processing; Color image segmentation processing: color image segmentation is performed on the spliced ​​orthophotos to separate different parts of the high-resolution images, where the different parts are grasslands and mouse holes; Grayscale processing: Convert color high-resolution images into grayscale images to simplify the process of high-resolution image processing; Image smoothing: Smoothing of high-resolution images; Edge detection processing: Use the Sobel operator to extract the edges of high-resolution images and detect edge information in high-resolution images.

4. The deep learning method for identifying rat holes in desertified grasslands based on a neural network model according to claim 3 is characterized in that: The preprocessing further comprises: Extracting a grayscale image, and traversing each pixel block in the grayscale image to obtain a grayscale value corresponding to each pixel block; According to the grayscale value of the pixel block in the central area of ​​the grayscale image, obtaining the central grayscale value corresponding to the grayscale image; Dividing the grayscale image according to a preset region size to obtain a plurality of image blocks; wherein the region size is n×n, and the value range of n is 3-9; Extract the central gray value corresponding to each image block; Obtaining a grayscale adjustment coefficient corresponding to each image block using the central grayscale value corresponding to each image block and the central grayscale value corresponding to the grayscale image; The grayscale adjustment coefficient is obtained by the following formula: Where S represents the grayscale adjustment coefficient; m represents the number of image blocks; H zi Represents the central grayscale value corresponding to the i-th image block; H z represents the central grayscale value corresponding to the grayscale image; x represents the adjustment coefficient, and the adjustment coefficient is obtained by the following formula: Where x represents the adjustment coefficient, m represents the number of image blocks, and H zi Represents the central grayscale value corresponding to the i-th image block; H z Indicates the central gray value corresponding to the gray image; H zmax and H zmin Represents the maximum and minimum values ​​of the central grayscale values ​​corresponding to the m image blocks; The grayscale value of the pixel block included in each image block is adjusted using the grayscale adjustment coefficient corresponding to each image block.

5. The deep learning method for identifying rat holes in desertified grasslands based on a neural network model according to claim 4 is characterized in that: The grayscale value of the pixel block included in each image block is adjusted by using the grayscale adjustment coefficient corresponding to each image block, including: Extracting the grayscale adjustment coefficient corresponding to each image block; Extracting the central grayscale value corresponding to each image block and the central grayscale value corresponding to the grayscale image; Obtaining a grayscale adjustment coefficient corresponding to the pixel blocks contained in each image block by using the grayscale value of each pixel block in each image block in combination with the central grayscale value of the corresponding image block and the central grayscale value corresponding to the grayscale image; The grayscale adjustment coefficient corresponding to the pixel block contained in each image block is obtained by the following formula: Wherein, F represents the grayscale adjustment coefficient corresponding to the pixel block contained in each image block; H represents the grayscale value corresponding to the pixel block contained in each image block; H zx Indicates the central gray value of the image block corresponding to the pixel block; H z Indicates the central gray value corresponding to the gray image; H zp Represents the average central grayscale value of all image blocks adjacent to the image block where the pixel block is located; The grayscale adjustment coefficient corresponding to each image block is combined with the grayscale adjustment coefficient corresponding to the pixel block contained in each image block to adjust the grayscale of the pixel block contained in each image block, and the grayscale value of the adjusted pixel block is obtained by the following formula: Among them, H t represents the gray value of the adjusted pixel block; F represents the gray adjustment coefficient corresponding to the pixel block contained in each image block; H represents the gray value corresponding to the pixel block contained in each image block; S represents the gray adjustment coefficient; ε represents the preset minimum constant to prevent the numerator from being 0.

6. The deep learning method for identifying rat holes in desertified grasslands based on a neural network model according to claim 1 is characterized in that: The identification of rat hole samples in desertified grasslands includes the following steps: Based on high-resolution images, spectral information of rat holes and their surroundings is obtained; Based on the spectral characteristics of rat holes and other grassland features in their surroundings, a rat hole index is proposed. According to the proposed rathole index, the rathole distribution image set is classified and identified to determine the location and number of ratholes; By identifying rat hole samples in desertified grasslands, a rat hole feature dataset was obtained.

7. The deep learning method for identifying rat holes in desertified grasslands based on a neural network model according to claim 6 is characterized in that: The rathole identification results are verified by comparing them with the real data on the ground and repeating the identification method. Once verified, the rathole feature data set is obtained and used to train the neural network model.

8. The deep learning method for identifying rat holes in desertified grasslands based on a neural network model according to claim 1 is characterized in that: The step 4 specifically includes the following steps: Model construction: Convolutional neural network is used as the basic model, and the network structure is designed according to the requirements of rathole identification tasks; Training model: Based on the neural network model, the rathole feature data set is used to train the neural network model, and then the neural network model is verified using the pre-prepared verification set to adjust the neural network model parameters and improve the performance of the neural network model; Model testing: Use the test set to test the trained neural network model and evaluate the accuracy of the neural network model.

9. The deep learning method for identifying rat holes in desertified grasslands based on a neural network model according to claim 8 is characterized in that: The training model comprises the following steps: Forward propagation: Forward propagation is performed during the training process. The characteristic variables of the input training samples are processed through each layer of the neural network model to finally obtain the output result. Calculate loss: Compare the output of the neural network model with the true label value and calculate the difference between the two; Back propagation: Perform the back propagation process and calculate the gradient of the loss function to the neural network model parameters based on the calculated loss through optimization algorithms such as gradient descent; Parameter update: Update the weight and bias parameters of the neural network model based on the calculated gradient information; Iterative optimization: The above steps will be iterated multiple times. Through multiple iterations, the neural network model gradually learns the mapping relationship from input to output.

10. The deep learning method for identifying rat holes in desertified grasslands based on a neural network model according to claim 8, characterized in that: The test set used for the model test is prepared in advance, and the data in the test set does not overlap with the rathole feature data set. Before conducting the model test, the test set is normalized and the processed test set is input into the neural network model to obtain the test results. According to the analysis results, the accuracy, precision and recall rate indicators of the neural network model are calculated.

Images