A method and system for automatic identification of cultivated land based on high-resolution images

Abstract

The present invention relates to an automatic identification method and system for cultivated land based on high-resolution images, which includes: preprocessing: dividing a vector file into different vector files, automatically merging multiple different raster files into a new raster file, and clipping the raster file with the vector file; cultivated land prediction: extracting plots with a network model, converting raster files to vector files, and converting line vector files to surface vector files; geometric element calculation: modifying the projection of the vector file, and calculating the geometric elements of the vector file; anomaly detection: bringing the ratio of the perimeter of the surface element to the area and the ratio of the area of ​​the surface element to the area of ​​the maximum rectangle into an anomaly detection model for anomaly detection, and finally obtaining a finely modified plot extraction file. The present invention has the advantages of high accuracy, high efficiency in identification and fine processing, and reduced labor costs, and has good versatility.

Description

Technical Field

[0001] The invention relates to the fields of image recognition, remote sensing application and agricultural planting structure, and in particular to a method for automatic recognition of cultivated land based on high-resolution images. Background Art

[0002] Cultivated land, as a kind of land used by humans to grow crops, is the most basic material that affects global agricultural production and food security. Its spatial distribution and quantity changes will have an impact on the world's food supply. In recent years, with the rapid development of the economy and the acceleration of urbanization and industrialization, the construction of roads and bridges, and the gradual increase in industrial and residential land have occupied more and more cultivated land, which is gradually exacerbating the global contradiction between human and land resources. In the face of increasing population and decreasing cultivated land, it is very important to obtain cultivated land information in a timely manner to deal with food crop production and safety issues. Therefore, in recent years, research on the identification and extraction of large-scale cultivated land information has attracted widespread attention. The early way to obtain cultivated land distribution and quantity information relied on manual field surveys, and the large amount of trivial data obtained was summarized and reported. Although this method can obtain high-precision cultivated land information data, it has obvious disadvantages. Relying on manual labor is not only time-consuming and labor-intensive, but also inefficient, and it is difficult to obtain real-time cultivated land data.

[0003] With the rapid development of science and technology, remote sensing data processing and artificial intelligence technology have been widely used. The use of high-resolution remote sensing image data has gradually become the most important means of efficiently extracting and analyzing cultivated land data. At present, combining the spectral information, geometric shape, and texture features in remote sensing data with various machine learning, deep learning and other classification algorithms to quickly identify and extract cultivated land is becoming a research hotspot. The use of supervised classification methods to quickly extract cultivated land information has been put on the agenda and has been applied to actual projects. For example, in order to develop a cultivated land protection system and a land conservation system, the third national land survey was conducted, and cultivated land identification based on high-resolution remote sensing and machine learning was used. However, due to the different geographical environments in which cultivated land is located, the different soil constitutions, the different crop types, and the different acquisition times of remote sensing images, the existing algorithms have poor recognition effects on cultivated land. Therefore, the current stage of cultivated land fine extraction still uses the more traditional manual visual interpretation method, which is very inefficient.

[0004] The existing methods for automatic identification, extraction and fine-tuning of cultivated land based on high-resolution images have complex and cumbersome data preprocessing, low efficiency, low recognition effect, problems of misidentification and missed identification, and are difficult to be directly applied in actual detection. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides a method and system for automatic identification of cultivated land based on high-resolution images to solve the problems of complex and cumbersome data preprocessing process, low efficiency and poor identification effect.

[0006] To achieve the above object, one aspect of the present invention provides a method for automatic identification of cultivated land based on high-resolution images, comprising the following steps:

[0007] S1. Preprocessing the first raster file and the first vector file to obtain a second vector file and a second raster file;

[0008] S2. Based on the second vector file and the second raster file, the cultivated land is extracted using a semantic segmentation network model to obtain a third raster file and convert it into a surface vector file;

[0009] S3. Convert the spatial data of the surface vector file into a coordinate system, and then calculate the geometric elements of the surface vector file;

[0010] S4. Based on the above-mentioned geometric elements, an anomaly detection model is trained, and an anomaly detection process is performed on the surface vector file to remove non-cultivated land vector elements to obtain a finely modified land plot extraction file.

[0011] Furthermore, in step S3, converting the spatial data of the surface vector file into a coordinate system includes:

[0012] The spatial data of the surface vector file is projected from the GCS_WGS_1984 geographic coordinate system to the AlbersCentral Asia geographic coordinate system.

[0013] Furthermore, the geometric elements include the area, perimeter, maximum horizontal width, maximum vertical height, maximum rectangular area, perimeter-area ratio α and ratio of the area of ​​the face element to the maximum rectangular area β of each face element.

[0014] Furthermore, the calculation formula of the perimeter-area ratio α is:

[0015] α = c / s, where c is the perimeter and s is the area;

[0016] The calculation formula for the ratio of the area of ​​the surface element to the area of ​​the largest rectangle β is:

[0017] β = s / smax, where smax is the maximum rectangular area.

[0018] Furthermore, the step S4 comprises:

[0019] S41. Divide the data set into a training set and a validation set, and obtain an anomaly detection model through training by a machine learning algorithm;

[0020] S42. Import the surface vector file data into the anomaly detection model and remove non-cultivated land surface elements.

[0021] Furthermore, the machine learning algorithm in S4 is a K-means clustering algorithm, an OCSVM support vector machine, a DBSCAN density clustering algorithm, a Bagging algorithm or an isolation forest algorithm.

[0022] Furthermore, the step S1 comprises:

[0023] S11. Splitting the input first vector file into a plurality of different second vector files;

[0024] S12. Merging a plurality of different first raster files to obtain a merged first raster file;

[0025] S13. According to the boundary of the second vector file, the merged first raster file is cropped to obtain a second raster file.

[0026] Furthermore, the step S2 comprises:

[0027] S21. Importing the second raster data and the second vector file into the neural network model for farmland extraction, obtaining a third raster file, and performing binarization processing;

[0028] S22. Using the first band of the third raster file as a mask layer, obtaining geographic reference information of the third raster file, and then automatically extracting features to convert the third raster file into a line vector file;

[0029] S23. For a closed area formed by one or more input elements, a new surface element is constructed and written into the output element file; and for input elements that cannot form a closed area, they are removed, thereby converting the line vector file into a surface vector file.

[0030] Furthermore, the semantic segmentation network model is D-LinkNet, Unet neural network or LinkNet34 neural network.

[0031] Another aspect of the present invention also provides a system for automatic identification of cultivated land based on high-resolution images, comprising: a computer-readable storage medium and a processor;

[0032] The computer-readable storage medium is used to store executable instructions;

[0033] The processor is used to read the executable instructions stored in the computer-readable storage medium to execute the above-mentioned method for automatic identification of cultivated land based on high-resolution images.

[0034] The beneficial effects brought about by the technical solution provided by the present invention include:

[0035] (1) The present invention utilizes a neural network model and Python automated batch processing technology to improve recognition accuracy while solving the problem of complex data preprocessing in the cultivated land recognition model, thereby improving the efficiency of batch recognition and extraction of cultivated land in high-resolution images, reducing labor costs, and realizing automation of cultivated land recognition.

[0036] (2) Combining the geometric shape rules of vector files with the random forest classification algorithm, a farmland refinement modification algorithm based on random forest is proposed, which further improves the accuracy of farmland identification and extraction based on high-resolution remote sensing images, and at the same time improves the efficiency of refined processing after farmland identification.

[0037] (3) The present invention is universal and can accurately and efficiently identify cultivated land in different application scenarios such as mountains, mid- and lower reaches, or deserts, and can also be applied to high-resolution image processing collected by different types of sensors. At the same time, it improves the efficiency of fine processing after cultivated land identification.

[0038] In general, the method for automatic identification, extraction and fine modification of cultivated land in high-resolution images proposed in the present invention first utilizes network models and python automated batch processing technology to improve the recognition accuracy while solving the problem of complex data preprocessing process of cultivated land identification model, thereby improving the efficiency of batch identification and extraction of cultivated land in high-resolution images; secondly, combining vector file geometric shape rules with random forest classification algorithm, a cultivated land fine modification algorithm based on random forest is proposed, which further improves the accuracy of cultivated land identification and extraction based on high-resolution remote sensing images, improves the efficiency of fine processing after cultivated land identification, greatly reduces labor costs, and realizes the automation and fine modification of cultivated land identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of automatic identification of cultivated land and refined modification of cultivated land in an embodiment of the present invention;

[0040] Figure 2 This is a segmentation effect diagram in an embodiment of the present invention;

[0041] Figure 3 It is the effect diagram before and after merging in the embodiment of the present invention;

[0042] Figure 4 This is a raster file clipping effect diagram in an embodiment of the present invention;

[0043] Figure 5 It is a raster file effect diagram predicted in an embodiment of the present invention;

[0044] Figure 6This is a line vector file effect diagram in an embodiment of the present invention;

[0045] Figure 7 This is a line vector file effect diagram in an embodiment of the present invention;

[0046] Figure 8 This is a vector file effect diagram in an embodiment of the present invention;

[0047] Fig. 9 This is a diagram showing the effect of the coordinate system projection transformation in an embodiment of the present invention;

[0048] Fig.10 The geometric element data of the vector file in the embodiment of the present invention;

[0049] Fig.11 It is a test of the model effect in the embodiment of the present invention;

[0050] Fig.12 This is a diagram showing the effect of abnormality detection in an embodiment of the present invention;

[0051] Fig.13 This is an overlay diagram of the anomaly detection effect and the raster file in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] 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.

[0053] The embodiment of the present invention provides a method and system for automatic identification of cultivated land based on high-resolution images. The present invention utilizes a network model and Python automated batch processing technology to improve the recognition accuracy while solving the problem of complex data preprocessing of the cultivated land identification model, thereby improving the efficiency of batch identification and extraction of cultivated land in high-resolution images; secondly, combining the geometric shape rules of vector files with the random forest classification algorithm, a cultivated land refinement modification algorithm based on random forest is proposed, which further improves the accuracy of cultivated land identification and extraction based on high-resolution remote sensing images, and at the same time improves the efficiency of refined processing after cultivated land identification.

[0054] Reference Figure 1 The embodiment of the present invention provides a method for automatic identification, extraction and fine modification of cultivated land based on high-resolution images, and the specific steps include:

[0055] Step 1: Automatically preprocess the remote sensing image raster files and vector files;

[0056] S11. Split the vector file into multiple different vector files: Extract each area element in the input vector file to form a separate output vector file. First, create an output vector file through the Writer() function in the shapefile library, then write the point set data of a surface element in the input vector file to the output vector file through the poly() function, and finally write the attribute value of the surface element to the output vector file using the record() function. The segmentation effect is as follows Figure 2 As shown, the dark vector file is a face feature segmentation file in the light vector file.

[0057] S12. Merge multiple different raster files into a new raster file: Based on the geographic coordinate data contained in the header of each raster file, use the MosaicToNewRaster_management() function in the arcpy library to automatically merge multiple small raster files into a new raster file according to the geographic coordinates of the raster files. The effect before and after the merger is shown in the figure below: Figure 3 As shown, the left side represents three unmerged raster files and the right side represents the merged raster file.

[0058] S13. Clipping of vector files to raster files: The merged raster files are clipped according to the boundaries of the vector files using the Clip_management() function in the arcpy library. The purpose is to reduce the amount of model data and the model prediction time. The effect of raster file clipping is shown in the figure below: Figure 4 shown.

[0059] Step 2: Cultivated land prediction

[0060] S21. D-LinkNet model to extract cultivated land: Use the load_state_dict() function in the networks library to read the trained neural network model, and then use the net.forward().squeeze().cpu() function in the torch library to predict the raster file, and finally binarize the image. The predicted raster file effect is shown in the figure below. Figure 5 shown.

[0061] In some embodiments, the neural network model may also be a Unet neural network and a LinkNet34 neural network.

[0062] S22. Convert raster files to vector files: First, use the GetRasterBand() function of the osgen library to get the first band of the raster file as the mask layer, then use the SpatialReference() function to get the georeference information of the raster file, and finally use the Polygonize() function to automatically extract features to convert the raster file to a vector file. The effect of the line vector file is shown below: Figure 6 shown.

[0063] S23. Convert line vector files to surface vector files: Use the FeatureToPolygon_management() function in the arcpy library to construct a new surface feature in the closed area formed by one or more input features and write it to the output feature file. For input features that cannot form a closed area, select to remove them. The effect of the line vector file is as follows Figure 7 As shown; the surface vector file effect diagram is as follows Figure 8 shown.

[0064] Step 3: Calculation of geometric elements

[0065] S31. Modify the vector file projection: Use Project_management() in the arcpy library to project spatial data from one coordinate system to another, and convert the GCS_WGS_1984 geographic coordinate system to the Albers CentralAsia geographic coordinate system. The coordinate system projection conversion effect diagram is as follows Fig. 9 shown.

[0066] S32. Calculate vector file geometry elements: vector file geometry elements are required when training samples and detecting cultivated land. The vector file geometry elements include the area, perimeter, maximum horizontal width, maximum vertical height, maximum rectangular area, perimeter-area ratio, and the ratio of the area of ​​the face element to the maximum rectangular area of ​​each face element in the face vector file.

[0067] Among them, the letter for area is s, the letter for perimeter is c, the letter for maximum horizontal width is w, the letter for maximum vertical height is h, and the letter for maximum rectangular area is smax.

[0068] The letter α for the perimeter-area ratio is calculated as:

[0069] α=c / s

[0070] The letter for the ratio of the area of ​​the face element to the area of ​​the largest rectangle is β, and its calculation formula is:

[0071] β=s / smax

[0072] The area is only related to the boundary of the cultivated land. The perimeter is used to distinguish between complex and simple geometric shapes. The maximum horizontal width and the maximum vertical height are used to calculate the area of ​​the rectangular boundary of the surface element. The ratio of the perimeter to the area is a parameter that describes the degree of wrinkling of the geometric shape. The larger the ratio, the higher the degree of wrinkling of the geometric shape per unit area.

[0073] The area of ​​each face feature is calculated by the CalculateAreas_stats() function in the arcpy library. The perimeter of each face feature is calculated by the CalculateField_management() function, and the maximum horizontal width, maximum vertical height, ratio of the perimeter to the area of ​​the face feature, and ratio of the area of ​​the face feature to the maximum rectangular area of ​​each face feature are calculated by the CalculateField_management() function. The geometric element data of the face vector file is as follows Fig.10 As shown, the geometric element data diagram is shown in Table 1.

[0074] Table 1

[0075]

[0076] Step 4: Anomaly Detection

[0077] S41. Training model: First, the random forest algorithm RandomForestClassifier() method in the sklearn library is used to train the random forest anomaly detection model using 90% of the training data set. Then, the remaining 10% of the training data set is used as the prediction indicator of the model to check the quality of the model. The model effect test is as follows: Fig.11 As shown, the test set accuracy can reach 93.8%.

[0078] In some embodiments, the anomaly detection model may also be an anomaly detection model trained by a bagging algorithm or an anomaly detection model trained by an isolation forest algorithm.

[0079] S42. Anomaly detection: First, use the load() function in the pickle library to read the trained model, then use the predict() function to predict the face vector file data, and finally write the face feature data corresponding to the predicted label into the new face vector file. The anomaly detection effect diagram is as follows Fig.12 As shown; the anomaly detection effect and the raster file overlay are shown in Fig.13 shown.

[0080] It should be noted that, in the present invention, relational terms such as "first" and "second" 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. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0081] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for automatic identification of cultivated land based on high-resolution images, characterized in that: The following steps are involved: S1. Preprocessing the first raster file and the first vector file to obtain a second vector file and a second raster file; S2. Based on the second vector file and the second raster file, the cultivated land is extracted using a semantic segmentation neural network model to obtain a third raster file and convert it into a surface vector file; S3. Convert the spatial data of the surface vector file into a coordinate system, and then calculate the geometric elements of the surface vector file; S4. Based on the above-mentioned geometric elements, an anomaly detection model is trained, and an anomaly detection process is performed on the surface vector file to remove non-cultivated land vector elements to obtain a finely modified land plot extraction file.

2. The method for automatic identification of cultivated land according to claim 1, characterized in that: In step S3, converting the spatial data of the surface vector file into a coordinate system includes: The spatial data of the surface vector file is projected from the GCS_WGS_1984 geographic coordinate system to the Albers CentralAsia geographic coordinate system.

3. The method for automatic identification of cultivated land according to claim 1 or 2, characterized in that: The geometric elements include the area, perimeter, maximum horizontal width, maximum vertical height, maximum rectangular area, perimeter-area ratio α and ratio of the area of ​​the face element to the maximum rectangular area β of each face element.

4. The method for automatic identification of cultivated land according to claim 3, characterized in that: The calculation formula of the perimeter-area ratio α is: α=c / s, where c is the perimeter and s is the area; The calculation formula for the ratio of the area of ​​the surface element to the area of ​​the largest rectangle β is: β=s / smax, where smax is the maximum rectangular area.

5. The method for automatic identification of cultivated land according to claim 1, characterized in that: The step S4 comprises: S41. Divide the data set into a training set and a validation set, and obtain an anomaly detection model through training by a machine learning algorithm; S42. Import the surface vector file data into the anomaly detection model and remove non-cultivated land surface elements.

6. The method for automatic identification of cultivated land according to claim 5, characterized in that: The machine learning algorithm in S4 is a K-means clustering algorithm, a single-class support vector machine algorithm, a DBSCAN clustering algorithm, a Bagging algorithm or an isolation forest algorithm.

7. The method for automatic identification of cultivated land according to claim 1, characterized in that: The step S1 comprises: S11. Splitting the input first vector file into a plurality of different second vector files; S12. Merging a plurality of different first raster files to obtain a merged first raster file; S13. According to the boundary of the second vector file, the merged first raster file is cropped to obtain a second raster file.

8. The method for automatic identification of cultivated land according to claim 1, characterized in that: The step S2 comprises: S21. Importing the second raster file and the second vector file into the semantic segmentation neural network model to extract cultivated land, obtaining a third raster file, and performing binarization processing; S22. Using the first band of the third raster file as a mask layer, obtaining geographic reference information of the third raster file, and then automatically extracting features to convert the third raster file into a line vector file; S23. For a closed area formed by one or more input elements, a new surface element is constructed and written into the output element file; and for input elements that cannot form a closed area, they are removed, thereby converting the line vector file into a surface vector file.

9. The method for automatic identification of cultivated land according to claim 8, characterized in that: The semantic segmentation neural network model is D-LinkNet, Unet neural network or LinkNet34 neural network.

10. An automatic farmland identification system based on high-resolution images, characterized in that: include: A computer readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is used to read the executable instructions stored in the computer-readable storage medium to execute the method for automatic identification of cultivated land based on high-resolution images as described in any one of claims 1 to 9.

Images