Farmland surface residual film pollution online monitoring method and system based on cloud platform

Abstract

The invention discloses a cloud platform-based farmland surface residual film pollution online monitoring method and system, and relates to the field of agricultural environment monitoring. The method comprises the following steps: deploying a trained earth surface residual film identification model on a cloud server; a surface residual film identification model adopts a U-Net model, the U-Net model is improved, and an improved cavity space pyramid pooling module, a convolution block attention module and an InceptionV4 network are introduced; using an unmanned aerial vehicle to collect a real-time farmland surface residual film image at a monitoring point position, and uploading the real-time farmland surface residual film image to a cloud server through a third-party cloud platform developed by a cloud API; on the cloud server, inputting a real-time farmland earth surface residual film image at a monitoring point into the trained earth surface residual film identification model, and outputting a farmland earth surface residual film identification result; and according to the farmland earth surface residual film identification result, determining the farmland earth surface residual film coverage rate, and further evaluating the farmland earth surface residual film pollution degree. According to the invention, online monitoring of farmland surface residual film pollution can be rapidly and accurately carried out.

Description

Technical Field

[0001] The present application relates to the field of agricultural environmental monitoring, and in particular to a cloud platform-based online monitoring method and system for residual film pollution on farmland surfaces. Background Art

[0002] Film mulching technology has the potential to increase warmth and moisture retention, reduce water loss, prevent pests and diseases, promote plant growth, and improve soil fertility. However, agricultural mulch films are commonly made of polyethylene. Due to overuse and a lack of timely recycling, the problem of residual agricultural mulch films is becoming increasingly serious. This leads to the annual accumulation of residual film in the soil, creating white pollution and posing a potential threat to agricultural production and the ecological environment. The treatment of residual film pollution in farmland is a comprehensive, systematic project. In addition to developing high-efficiency residual film recycling machinery, in-depth research on residual film pollution assessment is also crucial.

[0003] Currently, monitoring residual film pollution in farmland primarily relies on manual field sampling. However, this method suffers from low accuracy and efficiency, leading to delayed pollution monitoring and making it unsuitable for monitoring residual film pollution in large areas of farmland. Furthermore, current drone-based residual film pollution monitoring technology requires image acquisition followed by segmentation, a process that results in data lags. Summary of the Invention

[0004] The purpose of this application is to provide a cloud platform-based online monitoring method and system for residual film pollution on farmland surfaces, which can quickly and accurately monitor residual film pollution on farmland surfaces online.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In the first aspect, the present application provides an online monitoring method for residual film pollution on farmland surface based on a cloud platform, including: deploying a trained surface residual film recognition model on a cloud server; the surface residual film recognition model adopts a U-Net model, and introduces a void space pyramid pooling module between the encoder and the decoder of the U-Net model, and embeds a convolution block attention module after the convolution operation of the encoder of the U-Net model and after the void space pyramid pooling module, and integrates an InceptionV4 network before the convolution operation of the encoder of the U-Net model; using a drone to collect real-time residual film images of farmland surface at the monitoring point, and uploading them to the cloud server through a cloud API (Application Programming Interface); on the cloud server, inputting the real-time residual film images of farmland surface at the monitoring point into the trained surface residual film recognition model, and outputting the farmland surface residual film recognition result; determining the farmland surface residual film coverage rate based on the farmland surface residual film recognition result; and determining the degree of residual film pollution on the farmland surface based on the farmland surface residual film coverage rate.

[0007] Secondly, the present application provides a cloud-based online monitoring system for residual film pollution on farmland surfaces, comprising: a drone, a cloud API, and a cloud server. The drone is used to collect real-time images of residual film on the farmland surface at monitoring points and upload them to the cloud server via the cloud API; the cloud server is used to deploy a trained surface residual film recognition model, input the real-time images of residual film on the farmland surface at monitoring points into the trained surface residual film recognition model, and output the residual film recognition results; based on the residual film recognition results, the residual film coverage rate of the farmland surface is determined; and based on the residual film coverage rate of the farmland surface, the degree of residual film pollution on the farmland surface is determined.

[0008] According to the specific embodiments provided in this application, this application has the following technical effects:

[0009] The present application provides a cloud-based online monitoring method and system for residual film pollution on farmland surfaces. The void space pyramid pooling module can capture the feature information of multi-scale residual film images; the convolutional block attention module can highlight the residual film features in the extracted multi-scale information, reduce the impact of illumination, and suppress interference from irrelevant areas; the multi-branch structure of the InceptionV4 network can learn richer features, improve the feature extraction capability of the U-Net model under different lighting conditions, and enable the trained surface residual film recognition model to accurately identify residual film on farmland surfaces; through the low-altitude imaging technology of unmanned aerial vehicles and the cloud API, real-time collection and transmission of residual film images on farmland surfaces can be achieved, and combined with the trained surface residual film recognition model deployed on the cloud server, rapid and online monitoring of residual film on farmland surfaces can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] 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 any creative work.

[0011] Figure 1 A schematic diagram of a process for online monitoring of residual film pollution on farmland surfaces based on a cloud platform according to an embodiment of the present application;

[0012] Figure 2 A schematic diagram of the structure of an improved U-Net network provided in another embodiment of the present application;

[0013] Figure 3 A schematic structural diagram of an improved ASPP module provided in another embodiment of the present application;

[0014] Figure 4 A schematic diagram of data collection of residual film images on farmland surfaces provided by another embodiment of the present application;

[0015] Figure 5 A schematic diagram of misidentification results provided in another embodiment of the present application;

[0016] Figure 6 This is a schematic diagram of manual annotation of drone residual film images under different lighting conditions provided by another embodiment of the present application;

[0017] Figure 7 A schematic diagram of a fitting curve showing the predicted value and the actual value of the residual film coverage rate on the farmland surface provided by another embodiment of the present application;

[0018] Figure 8 A schematic diagram of the development strategy of a cloud-based online monitoring system for residual film pollution on farmland surfaces provided in one embodiment of the present application;

[0019] Figure 9 A schematic diagram of the integrated process of an online monitoring system for residual film pollution on farmland surfaces based on a cloud platform according to an embodiment of the present application;

[0020] Figure 10 This is a schematic diagram of the farmland residual film pollution detection report. DETAILED DESCRIPTION

[0021] 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.

[0022] 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.

[0023] In order to overcome the problems of low efficiency and poor accuracy in monitoring residual film pollution on farmland surfaces in the prior art, in an exemplary embodiment, Figure 1 As shown, a cloud platform-based online monitoring method for residual film pollution on farmland surfaces is provided, comprising the following steps 101 to 105. Among them:

[0024] Step 101: Deploy the trained surface film recognition model on the cloud server; the surface film recognition model adopts the U-Net model, and introduces a void space pyramid pooling module between the encoder and decoder of the U-Net model. After the convolution operation of the encoder of the U-Net model and after the void space pyramid pooling module, the convolution block attention module is embedded, and before the convolution operation of the encoder of the U-Net model, the InceptionV4 network is integrated.

[0025] Step 102: Use a drone to collect real-time images of residual film on the farmland surface at the monitoring point and upload them to the cloud server through the cloud API.

[0026] Step 103: On the cloud server, the real-time farmland surface film residue image at the monitoring point is input into the trained surface film residue recognition model, and the farmland surface film residue recognition result is output.

[0027] Step 104: Determine the coverage rate of residual film on the farmland surface based on the identification result of residual film on the farmland surface.

[0028] Step 105: Determine the pollution degree of residual film on the farmland surface according to the residual film coverage rate on the farmland surface.

[0029] Implement steps 101 to 105 above, and use drone low-altitude imaging technology and DJI cloud API to achieve real-time transmission and processing of residual film images on the farmland surface. Combined with the improved U-Net semantic segmentation model (i.e., the surface residual film recognition model), automatic recognition and segmentation of residual film can be achieved, thereby evaluating the degree of residual film pollution on the farmland surface.

[0030] In another exemplary embodiment of the present application, the surface film residual recognition model adopts the U-Net model, and the U-Net model is improved. The network structure of the improved U-Net model is as follows: Figure 2 As shown, the improvements are as follows:

[0031] (1) An improved Atrous Spatial Pyramid Pooling (ASPP) module is introduced in the transition layer between the encoder and decoder of the U-Net model, that is, between the upsampling layer and the downsampling layer, to expand the receptive field and thus capture the feature information of multi-scale residual film images.

[0032] The input of the improved ASPP module is a 32×64×1024 feature map, and the output is a 64×128×512 feature map that integrates information of different scales.

[0033] (2) After the convolution operation of the encoder and the improved ASPP module, a convolutional block attention module (CBAM) is embedded to highlight the residual film features in the extracted multi-scale information, reduce the influence of illumination, and suppress the interference of irrelevant areas.

[0034] The input of CBAM is 512×1024×64, 256×512×128, etc. The length and width of each feature map are half of the feature map of the previous layer, and the number of channels is doubled. The output is the feature map after spatial and channel enhancement, with the same dimensions as the input.

[0035] (3) Before the convolution operation of the encoder, the InceptionV4 network structure is integrated to improve the model's feature extraction ability under different lighting conditions. The multi-branch structure of the InceptionV4 network can learn richer features, and the improved ASPP module helps to introduce global context information into the network.

[0036] The first layer of the InceptionV4 network takes as input the preprocessed original image (a 512×1024×3 RGB image). The second layer takes as input the downsampled output of the first layer, arranged in order. The output of the InceptionV4 network is the probability of each category of background pixels and residual film pixels in the image at each downsampling step.

[0037] This application expands and improves the ASPP module, such as Figure 3 As shown in the figure, a layer of dilated convolution is added to the original structure, and a series of new dilation rates of 3, 6, 12, and 24 are adopted. The original ASPP is a three-layer convolution with dilation rates of 6, 12, and 18.

[0038] In another exemplary embodiment of the present application, the training process of the surface film residue recognition model can be replaced by the following steps 201 to 205:

[0039] Step 201: Using a drone to collect images of residual film on the farmland surface under different lighting conditions.

[0040] Step 202: each image of residual film on the farmland surface under different lighting conditions is labeled with residual film to obtain the labeled images of residual film on the farmland surface under different lighting conditions.

[0041] Exemplarily, Adobe Photoshop CS5 software is used to manually mark the residual film pixels.

[0042] Step 203: Upload the images of residual film on the farmland surface marked under different lighting conditions to the cloud server through the cloud API.

[0043] Step 204: Create a surface film residue identification model on the cloud server.

[0044] Step 205: using the farmland surface film residue images labeled under different lighting conditions to train the surface film residue recognition model to obtain a trained surface film residue recognition model.

[0045] In one example, step 201 may be replaced by the following steps 301 to 307 .

[0046] Step 301: Equip the DJI Mavic 3M drone with a visible light camera.

[0047] The visible light camera is a high-resolution visible light camera.

[0048] Step 302: During the flight of the DJI Mavic 3M drone, a visible light camera is used to collect multiple images of residual film on the farmland surface according to a preset collection time and a five-point sampling method.

[0049] The preset collection time is from 11:00 am to 19:00 pm, ensuring that images under different lighting conditions are collected, the ambient wind speed is maintained below level 3, the flight altitude is 5 meters, and the residual film on the surface has no obvious swing and the image clarity is guaranteed. Figure 4 shown.

[0050] Step 303: Classify the illumination of the image of residual film on the farmland surface collected on a cloudy day as weak light.

[0051] Step 304: Input the images of residual film on the farmland surface that are not collected on cloudy days into the trained U-Net model, perform semantic segmentation on the residual film on the farmland surface, and output the semantic segmentation results of the residual film on each image of residual film on the farmland surface.

[0052] Step 305: If misidentification occurs in the residual film semantic segmentation result, the illumination of the corresponding residual film image on the farmland surface is classified as exposure. Figure 5 Shows the misidentification result. Figure 5 Part (a) shows the first image of residual film on the farmland surface. Figure 5 Part (b) shows the misidentification result of the first farmland surface film residue image. Figure 5 Part (c) shows the second image of residual film on the farmland surface. Figure 5 Part (d) shows the misidentification result of the second image of residual film on the farmland surface.

[0053] Step 306: If there is no misidentification in the residual film semantic segmentation result, the lighting of the corresponding residual film image on the farmland surface is classified as normal light.

[0054] Step 307: The farmland surface residual film image under weak light, the farmland surface residual film image under exposure, and the farmland surface residual film image under normal light are combined to form farmland surface residual film images under different lighting conditions.

[0055] In another example, step 203 may be replaced by the following steps 401 to 403 .

[0056] Step 401: using a threshold segmentation method, performing a binarization operation on the farmland surface residual film images marked under different lighting conditions to obtain binarized farmland surface residual film images under different lighting conditions.

[0057] The residual film pixels in the binary farmland surface residual film image are marked as 1, and other background pixels such as soil are marked as 0, such as Figure 6 shown. Figure 6 Part (a) in the figure represents the exposure image. Figure 6 Part (b) in the figure represents the normal light image. Figure 6 Part (c) in the figure represents a low-light image. Figure 6 Part (d) in represents the exposure image label, Figure 6 Part (e) in represents the normal light image label, Figure 6 Part (f) in represents the low-light image label.

[0058] Step 402: Build a third-party cloud platform based on the cloud API.

[0059] Step 403: Using the online image transmission function of the third-party cloud platform, the binarized images of the residual film on the farmland surface under different lighting conditions are uploaded to the cloud server in real time.

[0060] In another example, the recognition accuracy of the improved U-Net model was trained and tested. The residual film coverage of farmland predicted by the improved U-Net model was linearly fitted with the true value. 2 Up to 0.968, such as Figure 7 The trained surface film recognition model is deployed to the cloud server.

[0061] In another exemplary embodiment of the present application, the image of residual film on the surface of farmland collected by the drone is input into a trained surface residual film recognition model to perform automatic recognition and segmentation of the residual film. The pixel area of ​​the residual film is calculated based on the segmentation result output by the model, and then the residual film coverage rate is obtained. For an image of residual film on the surface of farmland with a size of M×N, its residual film coverage rate L is the ratio of the total number of pixel points of the residual film (i.e., p(x,y)=1) to the total number of pixels in the image, that is, the residual film pixels divided by the total pixels are the residual film coverage rate. The specific implementation process of step 104 may include steps 501 and 502.

[0062] Step 501: Determine the pixel area of ​​the residual film based on the identification result of the residual film on the farmland surface.

[0063] Step 502: Based on the pixel area of ​​the residual film, use the formula Determine the residual film coverage rate on farmland surface.

[0064] Where L is the coverage rate of residual film on the farmland surface; p(x,y) is the pixel value of the residual film at the pixel point (x,y), and p(x,y)=1; x and y are the horizontal and vertical coordinates of the residual film pixel point, respectively; M is the number of pixels in the horizontal direction of the residual film image on the farmland surface, and N is the number of pixels in the vertical direction of the residual film image on the farmland surface.

[0065] In another exemplary embodiment of the present application, the pollution levels of residual film on the farmland surface include: clean, primary pollution, secondary pollution, tertiary pollution and quaternary pollution.

[0066] In another exemplary embodiment of the present application, a method for determining a value range for the degree of pollution of residual film on farmland surface includes steps 601 to 604.

[0067] Step 601: Count the surface residual film area and shallow residual film weight of different farmlands by mathematical statistics method.

[0068] For example, manual stratified sampling is performed, with a 1m×1m area dug to a depth of 30cm, and the weight of the residual film is calculated in layers.

[0069] Step 602: establishing a fitting regression prediction curve of the surface film residue area and the shallow film residue weight of the farmland according to the surface film residue area and the shallow film residue weight of the farmland of different farmlands.

[0070] Step 603: According to the standard range of the weight of the shallow residual film in the farmland under the degree of residual film pollution on the farmland surface, the value range of the surface residual film area under the degree of residual film pollution on the farmland surface is determined by using the fitting regression prediction curve.

[0071] Step 604: Determine the range of the surface residual film coverage rate under the degree of farmland surface residual film pollution according to the range of the surface residual film area under the degree of farmland surface residual film pollution, and form a range of the surface residual film coverage rate under the degree of farmland surface residual film pollution.

[0072] The method of this application uses the high-resolution camera onboard a DJI Mavic 3M drone to capture farmland surface images and transmits the image data in real time to a cloud server via the DJI cloud API. The cloud server deploys a deep learning model to perform semantic segmentation on the image, identifying and segmenting the residual film area. Based on the area ratio of the residual film area, the residual film coverage rate is calculated, and the level of residual film pollution in the farmland is assessed using mathematical statistics. This method can achieve rapid, accurate, and automated assessment of residual film pollution in farmland, providing technical support for farmland management and residual film pollution control.

[0073] Based on the same inventive concept, the embodiments of the present application also provide a cloud-based online monitoring system for residual film pollution on farmland surfaces, which is used to implement the aforementioned cloud-based online monitoring method for residual film pollution on farmland surfaces. The implementation solution provided by this system is similar to the implementation solution described in the aforementioned method. Therefore, the specific limitations of one or more cloud-based online monitoring system for residual film pollution on farmland surfaces provided below can be found in the above-mentioned limitations of the cloud-based online monitoring method for residual film pollution on farmland surfaces, and will not be repeated here.

[0074] In an exemplary embodiment, Figure 8 As shown, a cloud platform-based online monitoring system for residual film pollution on farmland surface is provided, including: a drone, a cloud API and a cloud server.

[0075] Drones are used to collect real-time images of residual film on farmland surfaces at monitoring points and upload them to a cloud server via a cloud API. The cloud server deploys a trained surface film recognition model, which inputs the real-time images of residual film on farmland surfaces from monitoring points into the trained model and outputs the residual film recognition results. Based on the recognition results, the residual film coverage rate is determined, and based on the residual film coverage rate, the degree of residual film contamination is determined.

[0076] For example, the images of residual film on the farmland surface collected by the drone are visible light images.

[0077] As an optional embodiment, the cloud server is further configured to create a surface film residue recognition model. The drone is further configured to collect images of residual film residue on farmland surfaces under different lighting conditions. The cloud server further uses the images of residual film residue on farmland surfaces under different lighting conditions to train the surface film residue recognition model, obtaining a trained surface film residue recognition model, and deploying the trained surface film residue recognition model.

[0078] Figure 9 This is a schematic diagram of the integration process for a cloud-based online monitoring system for residual film pollution on farmland surfaces. The integration process is as follows: drone image data collection of residual film on farmland → development of a third-party cloud platform based on a cloud API → construction of an improved U-Net semantic segmentation model → segmentation of residual film based on semantic segmentation → system integration for pollution monitoring.

[0079] Compared with traditional manual sampling methods, this application greatly reduces labor intensity, improves monitoring efficiency, and provides technical support for farmland management and residual film pollution control.

[0080] The workflow of the online monitoring system for residual film pollution on farmland surface based on the cloud platform is as follows:

[0081] (1) The DJI Mavic 3M, DJI third-party cloud platform, cloud server and improved U-Net semantic segmentation model were integrated to build an online monitoring system for residual film pollution on farmland surface.

[0082] (2) The system automatically collects images of residual film on the farmland surface through drones and transmits them to the cloud server in real time through the DJI cloud API.

[0083] (3) The images of the monitoring points are uploaded to the cloud server in real time. The residual film is identified by the improved U-Net semantic segmentation model deployed on the server. The residual film contamination rate is calculated based on the residual film pixel ratio. The residual film contamination level is evaluated based on the residual film contamination rate value. The monitoring results can be obtained within 5 minutes and a monitoring report can be issued.

[0084] The farmland film pollution detection report can show the farmer's name, plot address, plot latitude and longitude, plot area, film pollution monitoring results, time, etc. Figure 10 shown.

[0085] (4) Users can view the monitoring results of residual film pollution on farmland surface and check the flight trajectory and status of drones in real time through the web terminal.

[0086] DJI provides an MQTT-based cloud API built into Pilot 2. Using the built-in WebView engine, you can develop web pages based on your monitoring needs. These pages display system login, plot boundary selection, drone management, flight mission uploads, residual film pollution monitoring results, and residual film pollution monitoring report generation.

[0087] This application proposes a cloud-based online monitoring method and system for residual film pollution on farmland surfaces. This cloud platform enables real-time transmission, sharing, and visualization of monitoring data, enabling relevant personnel to obtain information on residual film pollution. Deep learning technology analyzes and processes farmland surface images to automatically identify and detect residual film, improving accuracy and efficiency. Autonomous drones, operating according to algorithms, can rapidly and comprehensively monitor farmland with far greater efficiency than manual monitoring. This application provides technical support for the real-time and rapid assessment of residual film pollution.

[0088] 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.

[0089] 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 cloud platform-based online monitoring method for residual film pollution on farmland surface, characterized in that: include: Deploy the trained surface film recognition model on a cloud server; the surface film recognition model uses a U-Net model and introduces a dilated spatial pyramid pooling module between the encoder and decoder of the U-Net model. A convolutional block attention module is embedded after the convolution operation of the U-Net model encoder and after the dilated spatial pyramid pooling module. An InceptionV4 network is integrated before the convolution operation of the U-Net model encoder. Use drones to collect real-time images of residual film on the farmland surface at monitoring points and upload them to the cloud server through the cloud API; On the cloud server, the real-time images of residual film on the farmland surface at the monitoring point are input into the trained residual film recognition model, and the residual film recognition results are output; According to the identification results of residual film on the farmland surface, determine the coverage rate of residual film on the farmland surface; The degree of pollution of residual film on the surface of farmland is determined based on the coverage rate of residual film on the surface of farmland.

2. The online monitoring method for residual film pollution on farmland surface based on cloud platform according to claim 1 is characterized in that: The training process of the surface film residual recognition model includes: Use drones to collect images of residual film on farmland surfaces under different lighting conditions; Each image of residual film on the farmland surface under different lighting conditions is labeled with residual film to obtain the labeled images of residual film on the farmland surface under different lighting conditions; Upload the images of residual film on the farmland surface marked under different lighting conditions to the cloud server through the cloud API; Create a surface film residue identification model on the cloud server; The surface film residue images of farmland labeled under different lighting conditions are used to train the surface film residue recognition model to obtain a trained surface film residue recognition model.

3. The online monitoring method for residual film pollution on farmland surface based on cloud platform according to claim 2 is characterized in that: The drone was used to collect images of residual film on the farmland surface under different lighting conditions, including: Equipped with a visible light camera on the DJI Mavic 3M drone; During the flight of the DJI Mavic 3M drone, a visible light camera was used to collect multiple images of residual film on the farmland surface according to the preset collection time and five-point sampling method; The illumination of the residual film images on the farmland surface collected on a cloudy day is classified as weak light; The images of residual film on the farmland surface that were not collected on cloudy days were input into the trained U-Net model to perform semantic segmentation on the residual film on the farmland surface, and the semantic segmentation results of the residual film on each image of the residual film on the farmland surface were output; If there is misidentification in the residual film semantic segmentation result, the illumination of the corresponding farmland surface residual film image is classified as exposure; If there is no misidentification in the residual film semantic segmentation result, the illumination of the corresponding farmland surface residual film image is classified as normal light; The images of residual film on the surface of farmland under weak light, the images of residual film on the surface of farmland under exposure and the images of residual film on the surface of farmland under normal light are combined to form images of residual film on the surface of farmland under different lighting conditions.

4. The online monitoring method for residual film pollution on farmland surface based on cloud platform according to claim 2 is characterized in that: Upload the images of residual film on the farmland surface marked under different lighting conditions to the cloud server through the cloud API, including: The threshold segmentation method is used to perform binarization operations on the images of residual film on the farmland surface marked under different lighting conditions to obtain the binarized images of residual film on the farmland surface under different lighting conditions. Build a third-party cloud platform based on cloud API; By utilizing the online image transmission function of a third-party cloud platform, the binary images of residual film on the farmland surface under different lighting conditions are uploaded to the cloud server in real time.

5. The online monitoring method for residual film pollution on farmland surface based on cloud platform according to claim 1 is characterized in that: The dilated spatial pyramid pooling module includes four layers of dilated convolutions, and the dilated ratios of each layer are 3, 6, 12, and 24, respectively.

6. The online monitoring method for residual film pollution on farmland surface based on cloud platform according to claim 1 is characterized in that: According to the identification results of residual film on the farmland surface, the coverage rate of residual film on the farmland surface is determined, including: According to the identification results of residual film on the farmland surface, the pixel area of ​​the residual film is determined; According to the pixel area of ​​the residual film, use the formula Determine the coverage rate of residual film on the farmland surface; where L is the coverage rate of residual film on the farmland surface; p(x,y) is the pixel value of the residual film at the pixel point (x,y), and p(x,y)=1; x and y are the horizontal and vertical coordinates of the residual film pixel point, respectively; M is the number of pixels in the horizontal direction of the residual film image on the farmland surface, and N is the number of pixels in the vertical direction of the residual film image on the farmland surface.

7. The online monitoring method for residual film pollution on farmland surface based on cloud platform according to claim 1 is characterized in that: The pollution levels of residual film on the farmland surface include: clean, primary pollution, secondary pollution, tertiary pollution and quaternary pollution.

8. The online monitoring method for residual film pollution on farmland surface based on cloud platform according to claim 1 or 7 is characterized in that: The method for determining the value range of the residual film pollution degree on farmland surface includes: By using mathematical statistics methods, the surface film residue area and shallow film residue weight of different farmlands were counted. According to the surface film residue area and shallow film residue weight of different farmlands, the fitting regression prediction curve of surface film residue area and shallow film residue weight of farmland was established; According to the standard range of the weight of the shallow residual film in the farmland under the degree of residual film pollution on the farmland surface, the value range of the surface residual film area under the degree of residual film pollution on the farmland surface is determined by using the fitting regression prediction curve; According to the value range of the surface residual film area under the degree of surface residual film pollution in farmland, the value range of the surface residual film coverage rate under the degree of surface residual film pollution in farmland is determined, forming the value range of the surface residual film coverage rate under the degree of surface residual film pollution in farmland.

9. An online monitoring system for residual film pollution on farmland surface based on cloud platform, characterized in that: include: Drones, cloud APIs, and cloud servers; Drones are used to collect real-time images of residual film on the farmland surface at monitoring points and upload them to the cloud server through the cloud API; The cloud server is used to deploy a trained surface residual film recognition model, input the real-time farmland surface residual film images at the monitoring points into the trained surface residual film recognition model, and output the farmland surface residual film recognition results; based on the farmland surface residual film recognition results, the farmland surface residual film coverage rate is determined; based on the farmland surface residual film coverage rate, the degree of farmland surface residual film pollution is determined.

10. The cloud platform-based online monitoring system for residual film pollution on farmland surface according to claim 9 is characterized in that: The cloud server is also used to create a surface film residual identification model; Drones are also used to collect images of residual film on farmland surfaces under different lighting conditions; The cloud server also uses the images of residual film on the surface of farmland under different lighting conditions to train the surface residual film recognition model, obtains the trained surface residual film recognition model, and deploys the trained surface residual film recognition model.

Images