A method and system for repairing residual film obstruction on farmland surface and predicting the unfolding area of curled residual film
By removing obstructions through an improved U-Net model and a fast frequency selective reconstruction algorithm, and combining it with a machine learning model to predict the expanded area of curled residual film, the problems of low accuracy in predicting the occlusion of residual film on the farmland surface and the expanded area of curled residual film were solved, achieving a more efficient residual film pollution assessment.
Patent Information
- Application Number
- CN202411868518.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-18
- Publication Date
- 2025-09-30
- Estimated Expiration
- 2044-12-18
AI Technical Summary
Existing technologies have low accuracy in predicting the area of residual film occlusion and curled residual film on farmland surfaces, especially under conditions of natural curling and object occlusion, making it difficult to achieve efficient and accurate residual film identification and pollution assessment.
An improved U-Net model is used for semantic segmentation, combined with a fast frequency selective reconstruction algorithm to remove occluders. The expanded area of the curled residual film is predicted through a machine learning model, and the curling index calculation formula is used for accurate prediction.
It improves the accuracy of repairing residual film obstructions on farmland surfaces, enhances the accuracy of predicting the unfolding area of curled residual film, and achieves more efficient residual film pollution assessment.
Smart Images
Figure CN119672010B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of machine vision technology, and in particular to a method and system for repairing residual film obstruction on farmland surface and predicting the unfolding area of curled residual film. Background Art
[0002] Agricultural mulch films offer excellent warming, moisture conservation, and weed control capabilities, effectively addressing the critical issue of extremely low yields in arid and cold regions, increasing yields by over 30%. They are a key tool for improving crop water use efficiency, alleviating water shortages, reducing pests and diseases, and promoting plant growth. However, currently, agricultural mulch films are primarily made of polyethylene, which has a long degradation period. Excessive use of these films and their failure to be recycled can lead to the accumulation of residual film in the soil, causing serious white pollution.
[0003] Managing residual film pollution in farmland is a systematic project. In addition to developing effective film recovery equipment, conducting research on residual film pollution assessment is also crucial. Pre-sowing residual film on farmland surfaces naturally curls and is obscured by objects, making it difficult to efficiently and accurately identify and assess the pollution.
[0004] Currently, the assessment and monitoring of residual film pollution in farmland is mostly carried out using manual field sampling, which is labor-intensive and inefficient. In recent years, researchers have conducted preliminary explorations into the feasibility of detecting residual film pollution on the ground using machine vision methods. To address the problem of assessing residual film pollution in farmland, Liang Changjiang's team conducted research on segmentation algorithms for residual film images using drones and conducted field film recognition tests. The iterative threshold segmentation algorithm achieved the best recognition accuracy, reaching 71%. Subsequently, to overcome the effects of illumination on residual film recognition accuracy, the team used the S component of images and a spiking neural network segmentation method to assess residual film pollution in tobacco fields after harvest, achieving an average recognition rate of 87.49%. However, this team's research focused on farmland that had not been treated with film removal or plowing after the autumn harvest, where the film was relatively continuous and had low fragmentation. Hu Can et al. used threshold segmentation on low-altitude drone images of residual film on the ground and used pixel area statistics to estimate the extent of residual film pollution, achieving an average recognition error of 4.82%. However, most of these studies used simple threshold segmentation or machine learning methods to directly segment and extract visible residual film fragments to assess the extent of residual film pollution on the ground.
[0005] Under the actual surface environmental conditions of farmland before sowing, the natural curling of residual film in the field and the obstruction of residual film by cotton stalks and drip irrigation tapes will have a great impact on the accuracy of residual film pollution assessment. Summary of the Invention
[0006] The purpose of this application is to provide a method and system for repairing residual film obstruction on farmland surface and predicting the area of curled residual film expansion, so as to solve the problem of low accuracy in repairing residual film obstruction on farmland surface and predicting the area of curled residual film expansion.
[0007] To achieve the above objectives, this application provides the following solutions:
[0008] In a first aspect, the present application provides a method for repairing residual film obstruction on farmland surface and predicting the unfolding area of curled residual film, comprising:
[0009] Acquire a target image; the target image is an image of the farmland to be tested when the curled residual film in the farmland to be tested is blocked by an obstruction;
[0010] Inputting the target image into a semantic segmentation model to obtain an image of the farmland to be tested with annotations; the annotations include: soil background, residual film, and obstructions; the semantic segmentation model is obtained by training an improved U-Net model using the first training set;
[0011] The pixels marked as soil background in the annotated image of the farmland to be tested are set to black, and the pixels marked as occlusions in the annotated image of the farmland to be tested are set to white, to obtain the image to be repaired;
[0012] Using a fast frequency selective reconstruction algorithm, the image to be repaired is subjected to occlusion removal and image repair to obtain a repaired image of the curled residual film; the repaired image of the curled residual film is a repaired image of the curled residual film in the farmland to be tested;
[0013] Determining curl index prediction parameters and pixel area of the curled residual film in the curled residual film repair image; the curl index prediction parameters include: color and texture features;
[0014] Inputting the curl index prediction parameters of the curled residual film of the tested farmland into the curl index prediction model to obtain a predicted value of the curl index of the curled residual film of the tested farmland; the curl index prediction model is obtained by training the machine learning model using the second training set;
[0015] The predicted value of the pixel area of the curled residual film after being unfolded in the tested farmland is determined according to the predicted value of the curling index of the curled residual film in the tested farmland and the pixel area of the curled residual film in the curled residual film repair image.
[0016] Optionally, the process of determining the semantic segmentation model includes:
[0017] Obtain a first training set; the first training set includes: images of a plurality of sample farmlands with curled residual film obscured by obstructions and corresponding annotated images;
[0018] Constructing the improved U-Net model; the improved U-Net model includes: an input module, multiple multi-scale feature extraction modules, multiple pooling modules, multiple convolution modules, multiple upsampling modules and an output module; the multi-scale feature extraction module includes a main branch of 3×3 standard convolution, an auxiliary branch of 1×1 convolution and an auxiliary branch of 5×5 convolution;
[0019] The improved U-Net model is trained with images of curled residual film of each sample farmland when it is blocked by an obstruction as input and corresponding labeled images as output to obtain the semantic segmentation model.
[0020] Optionally, using a fast frequency selective reconstruction algorithm to remove occluders and perform image restoration on the image to be restored, to obtain a restored image of the curled residual film, including:
[0021] Reconstructing each white pixel in the image to be repaired using a fast frequency selective reconstruction algorithm to obtain an initial repaired image; the initial repaired image includes black pixels and non-black pixels;
[0022] An area formed by non-black pixels in the initial repair image is determined as the curled residual film repair image.
[0023] Optionally, the process of determining the curl index prediction model includes:
[0024] Obtaining a second training set; the second training set includes: curl index prediction parameters and curl index measured values of a plurality of sample curled residual films;
[0025] Building the machine learning model;
[0026] The curl index prediction parameters of the curled residual films of each sample are used as input, and the corresponding measured values of the curl index are used as output, so as to train the machine learning model and obtain the curl index prediction model.
[0027] Optionally, determining a predicted value of a pixel area of the curled residual film after the curled residual film of the tested farmland is unfolded according to a predicted value of a curling index of the curled residual film of the tested farmland and a pixel area of the curled residual film in a curled residual film repair image includes:
[0028] The curl index calculation formula is used to calculate the predicted value of the pixel area of the curled residual film after it is unfolded, based on the predicted value of the curl index of the curled residual film of the tested farmland and the pixel area of the curled residual film in the curled residual film repair image. The curl index calculation formula is:
[0029]
[0030] Wherein, RFCI is the curl index; S1 is the pixel area of the curled residual film; and S2 is the pixel area after the curled residual film is unfolded.
[0031] Optionally, the curl index calculation formula is obtained by fitting the measured values of the pixel areas of a plurality of sample curled residual films, the measured values of the pixel areas after unfolding, and the measured values of the curl index.
[0032] In a second aspect, the present application provides a system for repairing residual film obstructions on farmland surfaces and predicting the area of curled residual film expansion, so as to implement any of the above-mentioned methods for repairing residual film obstructions on farmland surfaces and predicting the area of curled residual film expansion. The system for repairing residual film obstructions on farmland surfaces and predicting the area of curled residual film expansion comprises:
[0033] A target image acquisition module is used to acquire a target image; the target image is an image of the farmland to be tested when the curled residual film in the farmland to be tested is blocked by an obstruction;
[0034] A semantic segmentation module is configured to input the target image into a semantic segmentation model to obtain an annotated image of the farmland to be tested; the annotated images include soil background, residual film, and occluders; the semantic segmentation model is obtained by training an improved U-Net model using the first training set;
[0035] The module for determining the image to be repaired is used to set all pixels marked as soil background in the annotated image of the farmland to be tested to black, and to set all pixels marked as occlusions in the annotated image of the farmland to be tested to white, thereby obtaining the image to be repaired;
[0036] A residual film repair module is used to remove occluders and perform image repair on the image to be repaired using a fast frequency selective reconstruction algorithm to obtain a repaired image of the curled residual film; the repaired image of the curled residual film in the farmland to be tested is a repaired image;
[0037] A parameter acquisition module is used to determine the curl index prediction parameters and pixel area of the curled residual film in the curled residual film repair image; the curl index prediction parameters include: color and texture features;
[0038] a curl index prediction module, configured to input curl index prediction parameters of the curled residual film of the tested farmland into a curl index prediction model to obtain a predicted value of the curl index of the curled residual film of the tested farmland; the curl index prediction model is obtained by training a machine learning model using the second training set;
[0039] The unfolding area prediction module is used to determine the predicted value of the pixel area of the curled residual film in the farmland to be tested after it is unfolded based on the predicted value of the curling index of the curled residual film in the farmland to be tested and the pixel area of the curled residual film in the curled residual film repair image.
[0040] According to the specific embodiments provided in this application, this application discloses the following technical effects:
[0041] The present application discloses a method and system for repairing occlusion of residual film on the surface of farmland and predicting the expansion area of curled residual film. First, a target image is obtained; the target image is an image of the farmland to be tested when the curled residual film in the farmland to be tested is blocked by an obstruction; then, the target image is input into a semantic segmentation model to obtain an annotated image of the farmland to be tested; the annotations include: soil background, residual film and obstruction; the semantic segmentation model is obtained by training an improved U-Net model using a first training set; secondly, each pixel marked as soil background in the annotated image of the farmland to be tested is set to black, and each pixel marked as obstruction in the annotated image of the farmland to be tested is set to white to obtain an image to be repaired; then, a fast frequency selection is used to select the target image. A selective reconstruction algorithm is used to remove occluders and perform image restoration on the image to be repaired to obtain a repaired image of curled residual film; curl index prediction parameters and pixel area of the curled residual film in the repaired image of curled residual film are determined; curl index prediction parameters include: color and texture features; again, the curl index prediction parameters of the curled residual film of the farmland to be tested are input into the curl index prediction model to obtain the predicted value of the curl index of the curled residual film of the farmland to be tested; the curl index prediction model is obtained by training the machine learning model using the second training set; finally, according to the predicted value of the curl index of the curled residual film of the farmland to be tested and the pixel area of the curled residual film in the repaired image of curled residual film, the predicted value of the pixel area of the curled residual film of the farmland to be tested after being unfolded is determined. This application realizes the repair of occluded residual film through multi-classification semantic segmentation and image restoration methods, and realizes the prediction of the pixel area of the curled residual film after being unfolded through machine learning, thereby realizing the repair of occluded residual film on the farmland surface and improving the prediction accuracy of the unfolded area of curled residual film. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0043] Figure 1 A schematic flow chart of a method for repairing residual film obstruction on farmland surface and predicting the unfolded area of curled residual film provided in one embodiment of the present application;
[0044] Figure 2 This is a schematic diagram of an image of the farmland to be tested when the curled residual film is blocked by an obstruction;
[0045] Figure 3This is a schematic diagram of the image of the curled residual film in the sample farmland when it is blocked by an obstruction;
[0046] Figure 4 Manually annotated schematic diagrams;
[0047] Figure 5 Schematic diagram of the visualization results of the annotated image;
[0048] Figure 6 To improve the U-Net model structure diagram;
[0049] Figure 7 This is a schematic diagram of the multi-scale feature extraction module structure;
[0050] Figure 8 Schematic diagram of an image with annotations obtained by the semantic segmentation model;
[0051] Figure 9 is a schematic diagram of the image to be repaired;
[0052] Figure 10 Schematic diagram of the image repair of curled residual film;
[0053] Figure 11 This is a schematic diagram of the curled residual film of the sample;
[0054] Figure 12 A schematic diagram of the pixel area of the curled residual film in the curled residual film repair image;
[0055] Figure 13 This is a schematic diagram of the image of the curled residual film in the unfolding state;
[0056] Figure 14 Schematic diagram of the pixel area after the curled residual film is unfolded. DETAILED DESCRIPTION
[0057] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.
[0058] The purpose of this application is to provide a method and system for repairing residual film obstruction on the surface of farmland and predicting the area of curled residual film expansion, aiming to achieve repair of residual film obstruction on the surface of farmland and improve the prediction accuracy of the area of curled residual film expansion.
[0059] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.
[0060] In an exemplary embodiment, Figure 1 As shown, a method for repairing residual film obstruction on farmland surface and predicting the area of curled residual film expansion is provided, including:
[0061] Step 1: Acquire a target image; the target image is an image of the farmland to be tested when the curled residual film in the farmland to be tested is blocked by an obstruction.
[0062] Specifically, the collected image of the farmland to be tested when the curled residual film is blocked by an obstruction is as follows: Figure 2 shown.
[0063] Step 2: Input the target image into the semantic segmentation model to obtain an annotated image of the farmland to be tested.
[0064] The annotations include: soil background, residual film and occlusions; the semantic segmentation model is obtained by training the improved U-Net model using the first training set.
[0065] As an optional implementation, the process of determining the semantic segmentation model includes:
[0066] Step 21: Obtain a first training set; the first training set includes: images of a plurality of sample farmlands with curled residual film blocked by obstructions and corresponding annotated images.
[0067] Specifically, a 1m×1m sample box was used to lay out a residual film sample collection area (i.e., multiple sample farmlands). A digital camera was used to collect images of the curled residual film in multiple sample farmlands when it was obscured by obstructions such as straw, drip irrigation tape, cotton, and boll shells. The image of the curled residual film in one of the sample farmlands when it was obscured by obstructions is shown in the figure below. Figure 3 As shown, using Figure 4 After manual annotation, we get Figure 5 The corresponding visualization results with annotations are shown.
[0068] Step 22: Construct an improved U-Net model; the improved U-Net model includes: an input module, multiple multi-scale feature extraction modules, multiple pooling modules, multiple convolution modules, multiple upsampling modules and an output module; the multi-scale feature extraction module includes a main branch of 3×3 standard convolution, an auxiliary branch of 1×1 convolution and an auxiliary branch of 5×5 convolution.
[0069] Specifically, such as Figure 6As shown in the figure, the improved U-Net model uses the classic semantic segmentation network U-Net framework. It replaces the 3×3 standard convolution during the downsampling process with a multi-scale feature extraction module to extract multi-scale information about the residual film and its occluders. Furthermore, to further speed up the model and reduce the amount of computation, compared to the original U-Net network, the improved U-Net model removes one convolutional layer during both the downsampling and upsampling processes, making the model more lightweight.
[0070] like Figure 7 As shown in the figure, the main branch of the multi-scale feature extraction module is a 3×3 standard convolution, with 1×1 and 5×5 convolutions added as auxiliary branches. This allows the extraction of 3×3 image features as well as 1×1 and 5×5 image features, enhancing image feature extraction capabilities, improving the model's utilization of image information, broadening the nonlinear layer, and strengthening nonlinear expression capabilities. Finally, the outputs of the three branches, namely the 1×1 convolution, the 3×3 standard convolution, and the 5×5 convolution, are fused using a 1×1 convolution.
[0071] Step 23: Using the image of the curled residual film of each sample farmland when it is blocked by an obstruction as input and the corresponding labeled image as output, the improved U-Net model is trained to obtain a semantic segmentation model.
[0072] Specifically, after the training is completed, the semantic segmentation model is used to Figure 3 The image of the curled residual film of the sample farmland shown in the figure is blocked by an obstruction and the resulting labeled image is as follows: Figure 8 shown.
[0073] Step 3: Set all pixels marked as soil background in the annotated image of the farmland to be tested to black, and set all pixels marked as occlusions in the annotated image of the farmland to be tested to white, to obtain the image to be repaired.
[0074] Specifically, the RGB values of all pixels marked as soil background in the labeled image of the farmland to be tested are set to (0, 0, 0), so that the corresponding pixels are set to black, and the RGB values of all pixels marked as occlusion in the labeled image of the farmland to be tested are set to (256, 256, 256), so that the corresponding pixels are set to white, and the image to be repaired is obtained. Figure 2 The image to be repaired corresponds to the image of the farmland to be tested when the curled residual film is blocked by an obstruction. Figure 9 shown.
[0075] Step 4: Use a fast frequency selective reconstruction algorithm to remove occluders and perform image restoration on the image to be restored, thereby obtaining a restored image of the curled residual film; the restored image of the curled residual film is the restored image of the curled residual film in the farmland to be tested.
[0076] As an optional implementation, step 4 includes:
[0077] Step 41: Reconstruct each white pixel in the image to be repaired using a fast frequency selective reconstruction algorithm to obtain an initial repaired image; the initial repaired image includes black pixels and non-black pixels.
[0078] Step 42: Determine the area formed by the non-black pixels in the initial repair image as the curled residual film repair image.
[0079] Specifically, Figure 2 The curled residual film repair image corresponding to the image of the farmland to be tested when the curled residual film is blocked by an obstruction is as shown in FIG. Figure 10 shown.
[0080] Step 5: Determine the curl index prediction parameters and pixel area of the curled residual film in the curled residual film repair image; the curl index prediction parameters include: color and texture features.
[0081] Step 6: Input the curl index prediction parameters of the curled residual film of the farmland to be tested into the curl index prediction model to obtain the predicted value of the curl index of the curled residual film of the farmland to be tested.
[0082] Among them, the curl index prediction model is obtained by training the machine learning model using the second training set.
[0083] As an optional implementation, the process of determining the curl index prediction model includes:
[0084] Step 61: Obtain a second training set; the second training set includes: curl index prediction parameters and actual values of curl index of a plurality of sample curled residual films.
[0085] Step 62: Build a machine learning model.
[0086] Step 63: Using the curl index prediction parameters of each sample curled residual film as input and the corresponding measured value of the curl index as output, the machine learning model is trained to obtain a curl index prediction model.
[0087] Step 7: Determine the predicted value of the pixel area of the curled residual film after it is unfolded in the tested farmland based on the predicted value of the curling index of the curled residual film in the tested farmland and the pixel area of the curled residual film in the curled residual film repair image.
[0088] As an optional implementation, step 7 includes:
[0089] The curl index calculation formula is used to calculate the predicted value of the pixel area of the curled residual film after it is unfolded, based on the predicted value of the curl index of the curled residual film in the tested farmland and the pixel area of the curled residual film in the curled residual film repair image. The curl index calculation formula is:
[0090]
[0091] Wherein, RFCI is the curl index; S1 is the pixel area of the curled residual film; and S2 is the pixel area after the curled residual film is unfolded.
[0092] As an optional implementation, the curl index calculation formula is obtained by fitting the measured values of the pixel areas of multiple sample curled residual films, the measured values of the pixel areas after unfolding, and the measured values of the curl index.
[0093] Specifically, when fitting the curl index calculation formula, one of the images of the curled residual film of the sample is as follows Figure 11 As shown, Figure 11 Perform semantic segmentation model to obtain the corresponding labeled image. Figure 11 The corresponding pixels marked as soil background in the annotated image are set to black. Figure 11 The corresponding pixels marked as occluders in the annotated image are set to white, and the obtained Figure 11 The corresponding image to be repaired is reconstructed using a fast frequency selective reconstruction algorithm. Figure 11 The corresponding image to be repaired is subjected to occlusion removal and image repair, and the obtained Figure 11 The corresponding curled residual film repair image is used to determine Figure 12 shown Figure 11 The pixel area of the curled residual film in the corresponding curled residual film repair image. Figure 12 The pixel area in is 11976Pixel (i.e. 11976 pixels). Figure 11 After the curled residual film of the sample is cleaned and unfolded, the image of the residual film unfolding state is as follows Figure 13 As shown, the Otsu threshold segmentation algorithm is used to segment the residual film expansion state image to determine Figure 14 The pixel area of the sample after the curled residual film is unfolded is shown. Figure 14 The pixel area is 30226Pixel (i.e. 30226 pixels).
[0094] In an exemplary embodiment, a system for repairing residual film obstruction on a farmland surface and predicting the area of curled residual film expansion is provided to implement any of the above-mentioned methods for repairing residual film obstruction on a farmland surface and predicting the area of curled residual film expansion. The system for repairing residual film obstruction on a farmland surface and predicting the area of curled residual film expansion includes:
[0095] The target image acquisition module is used to obtain a target image; the target image is an image of the farmland to be tested when the curled residual film in the farmland to be tested is blocked by an obstruction.
[0096] The semantic segmentation module is used to input the target image into the semantic segmentation model to obtain an annotated image of the farmland to be tested; the annotations include: soil background, residual film and occlusions; the semantic segmentation model is obtained by training the improved U-Net model using the first training set.
[0097] The module for determining the image to be repaired is used to set all pixels marked as soil background in the annotated image of the farmland to be tested to black, and to set all pixels marked as occlusions in the annotated image of the farmland to be tested to white, so as to obtain the image to be repaired.
[0098] The residual film repair module is used to remove occlusions and perform image repair on the image to be repaired using a fast frequency selective reconstruction algorithm to obtain a curled residual film repair image; the curled residual film repair image is an image of the curled residual film in the farmland to be tested after repair.
[0099] The parameter acquisition module is used to determine the curl index prediction parameters and pixel area of the curled residual film in the curled residual film repair image; the curl index prediction parameters include: color and texture features.
[0100] The curling index prediction module is used to input the curling index prediction parameters of the curled residual film of the farmland to be tested into the curling index prediction model to obtain the predicted value of the curling index of the curled residual film of the farmland to be tested; the curling index prediction model is obtained by training the machine learning model using the second training set.
[0101] The unfolding area prediction module is used to determine the predicted value of the pixel area of the curled residual film in the farmland to be tested after it is unfolded based on the predicted value of the curling index of the curled residual film in the farmland to be tested and the pixel area of the curled residual film in the curled residual film repair image.
[0102] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0103] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.
Claims
1. A method for repairing residual film obstruction on farmland surface and predicting the unfolding area of curled residual film, characterized in that: The method for repairing residual film obstruction on the farmland surface and predicting the area of curled residual film expansion includes: Acquire a target image; the target image is an image of the farmland to be tested when the curled residual film in the farmland to be tested is blocked by an obstruction; Inputting the target image into a semantic segmentation model to obtain an image of the farmland to be tested with annotations; the annotations include: soil background, residual film, and obstructions; the semantic segmentation model is obtained by training an improved U-Net model using the first training set; The pixels marked as soil background in the annotated image of the farmland to be tested are set to black, and the pixels marked as occlusions in the annotated image of the farmland to be tested are set to white, to obtain the image to be repaired; Using a fast frequency selective reconstruction algorithm, the image to be repaired is subjected to occlusion removal and image repair to obtain a repaired image of the curled residual film; the repaired image of the curled residual film is a repaired image of the curled residual film in the farmland to be tested; Determining curl index prediction parameters and pixel area of the curled residual film in the curled residual film repair image; the curl index prediction parameters include: color and texture features; Inputting the curl index prediction parameters of the curled residual film of the tested farmland into the curl index prediction model to obtain a predicted value of the curl index of the curled residual film of the tested farmland; the curl index prediction model is obtained by training the machine learning model using the second training set; The predicted value of the pixel area of the curled residual film after being unfolded in the tested farmland is determined according to the predicted value of the curling index of the curled residual film in the tested farmland and the pixel area of the curled residual film in the curled residual film repair image.
2. The method for repairing residual film obstruction and predicting the expansion area of curled residual film on farmland surface according to claim 1 is characterized in that: The process of determining the semantic segmentation model includes: Obtain a first training set; the first training set includes: images of a plurality of sample farmlands with curled residual film obscured by obstructions and corresponding annotated images; Constructing the improved U-Net model; the improved U-Net model includes: an input module, multiple multi-scale feature extraction modules, multiple pooling modules, multiple convolution modules, multiple upsampling modules and an output module; the multi-scale feature extraction module includes a main branch of 3×3 standard convolution, an auxiliary branch of 1×1 convolution and an auxiliary branch of 5×5 convolution; The improved U-Net model is trained with images of curled residual film of each sample farmland when it is blocked by an obstruction as input and corresponding labeled images as output to obtain the semantic segmentation model.
3. The method for repairing residual film obstruction and predicting the expansion area of curled residual film on farmland surface according to claim 1 is characterized in that: Using a fast frequency selective reconstruction algorithm, the image to be repaired is subjected to occlusion removal and image repair to obtain a repaired image of the curled residual film, including: Reconstructing each white pixel in the image to be repaired using a fast frequency selective reconstruction algorithm to obtain an initial repaired image; the initial repaired image includes black pixels and non-black pixels; An area formed by non-black pixels in the initial repair image is determined as the curled residual film repair image.
4. The method for repairing residual film obstruction and predicting the expansion area of curled residual film on farmland surface according to claim 1, characterized in that: The process of determining the curl index prediction model includes: Obtaining a second training set; the second training set includes: curl index prediction parameters and curl index measured values of a plurality of sample curled residual films; Building the machine learning model; The curl index prediction parameters of the curled residual films of each sample are used as input, and the corresponding measured values of the curl index are used as output, so as to train the machine learning model and obtain the curl index prediction model.
5. The method for repairing residual film obstruction and predicting the expansion area of curled residual film on farmland surface according to claim 1 is characterized in that: Determining a predicted value of the pixel area of the curled residual film after the curled residual film of the tested farmland is unfolded according to the predicted value of the curled residual film of the tested farmland and the pixel area of the curled residual film in the curled residual film repair image, including: The curl index calculation formula is used to calculate the predicted value of the pixel area of the curled residual film after it is unfolded, based on the predicted value of the curl index of the curled residual film of the tested farmland and the pixel area of the curled residual film in the curled residual film repair image. The curl index calculation formula is: Wherein, RFCI is the curl index; S1 is the pixel area of the curled residual film; and S2 is the pixel area after the curled residual film is unfolded.
6. The method for repairing residual film obstruction and predicting the expansion area of curled residual film on farmland surface according to claim 5, characterized in that: The curl index calculation formula is obtained by fitting the measured values of the pixel areas of a plurality of sample curled residual films, the measured values of the pixel areas after unfolding, and the measured values of the curl index.
7. A system for repairing residual film on farmland surface and predicting the area of curled residual film expansion, to implement the method for repairing residual film on farmland surface and predicting the area of curled residual film expansion as described in any one of claims 1 to 6, characterized in that: The farmland surface residual film obstruction repair and curled residual film expansion area prediction system includes: A target image acquisition module is used to acquire a target image; the target image is an image of the farmland to be tested when the curled residual film in the farmland to be tested is blocked by an obstruction; A semantic segmentation module is configured to input the target image into a semantic segmentation model to obtain an annotated image of the farmland to be tested; the annotated images include soil background, residual film, and occluders; the semantic segmentation model is obtained by training an improved U-Net model using the first training set; The module for determining the image to be repaired is used to set all pixels marked as soil background in the annotated image of the farmland to be tested to black, and to set all pixels marked as occlusions in the annotated image of the farmland to be tested to white, thereby obtaining the image to be repaired; A residual film repair module is used to remove occluders and perform image repair on the image to be repaired using a fast frequency selective reconstruction algorithm to obtain a repaired image of the curled residual film; the repaired image of the curled residual film in the farmland to be tested is a repaired image; A parameter acquisition module is used to determine the curl index prediction parameters and pixel area of the curled residual film in the curled residual film repair image; the curl index prediction parameters include: color and texture features; a curl index prediction module, configured to input curl index prediction parameters of the curled residual film of the tested farmland into a curl index prediction model to obtain a predicted value of the curl index of the curled residual film of the tested farmland; the curl index prediction model is obtained by training a machine learning model using the second training set; The unfolding area prediction module is used to determine the predicted value of the pixel area of the curled residual film in the farmland to be tested after it is unfolded based on the predicted value of the curling index of the curled residual film in the farmland to be tested and the pixel area of the curled residual film in the curled residual film repair image.