Farmland construction quality intelligent sampling inspection method, system and device and storage medium

Abstract

The invention discloses an intelligent spot check method, system and device for farmland construction quality and a storage medium, and relates to the technical field of agricultural monitoring. The method comprises the following steps: acquiring remote sensing environment data of a target area; the remote sensing environment data comprises a multi-temporal remote sensing image and ground environment data; inputting the remote sensing environment data into a high-standard farmland time sequence collaborative sampling model, and determining a spot check prediction result; the prediction result comprises a risk thermodynamic diagram and an uncertainty quantization map; the high-standard farmland time sequence collaborative sampling model is constructed based on a variation function model, a time sequence triggering mechanism and a Monte Carlo technology, and is used for dynamically optimizing a sampling strategy; and performing uncertainty analysis based on the spot check prediction result to generate a final spot check area. According to the invention, global precise monitoring and dynamic optimization regulation and control of high-standard farmland construction quality can be realized, and technical support is provided for long-term guarantee of farmland quality in black soil areas.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural monitoring, and particularly to an intelligent sampling inspection method, system, device and storage medium for the quality of farmland construction. Background Art

[0002] The quality of high-standard farmland construction is a key factor in ensuring food safety production and enhancing the agricultural disaster prevention and mitigation ability, and is of great significance for realizing agricultural modernization and sustainable development. However, the current quality inspection system still mainly relies on traditional means (such as manual visual inspection, tape measure measurement, core drilling measurement, etc.). These methods are inefficient, costly, and difficult to achieve rapid and comprehensive monitoring of construction quality, and cannot meet the detection requirements of large-scale and high-precision.

[0003] With the rapid development of multi-source remote sensing technology and deep learning technology, these advanced technologies have been initially applied in the large-scale and rapid sampling inspection of high-standard farmland. However, the existing technologies are mostly limited to a single data source or a single construction content (such as farmland planning, irrigation facilities, shelter forest construction, etc.), and fail to comprehensively cover multi-dimensional elements such as farmland infrastructure, crop growth conditions, soil quality, meteorological conditions and disaster impacts, resulting in insufficient representativeness and comprehensiveness of the sampling inspection results. In addition, the existing methods lack the ability to dynamically monitor the entire life cycle of high-standard farmland construction (such as the construction period, management and protection period) and the critical phenological periods of crops (such as land preparation before sowing, soil restoration after harvest), and it is difficult to achieve the temporal coordination and optimization of construction quality. At the same time, the existing technologies have limited ability to analyze spatial heterogeneity, especially in complex terrain or areas with uneven data distribution, and the prediction accuracy and reliability need to be improved urgently. These limitations seriously restrict the scientific evaluation and effective supervision of the quality of high-standard farmland construction. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent sampling inspection method, system, device and storage medium for the quality of farmland construction, which can achieve all-region precise monitoring and dynamic optimization and regulation of the quality of high-standard farmland construction, and provide technical support for the long-term guarantee of farmland quality in black soil areas.

[0005] To achieve the above purpose, the present invention provides the following solutions:

[0006] An intelligent sampling inspection method for the quality of farmland construction, comprising:

[0007] Obtaining remote sensing environmental data of a target area; the remote sensing environmental data includes multi-temporal remote sensing images and ground environmental data;

[0008] Input the remote sensing environmental data into the high-standard farmland time-series collaborative sampling model to determine the sampling inspection prediction results; the prediction results include a risk heat map and an uncertainty quantification map; the high-standard farmland time-series collaborative sampling model is constructed based on a variogram model, a time-series triggering mechanism, and Monte Carlo techniques for dynamically optimizing the sampling strategy;

[0009] Conduct uncertainty analysis based on the sampling inspection prediction results to generate the final sampling inspection area.

[0010] Optionally, the process of obtaining the remote sensing environmental data of the target area specifically includes:

[0011] Obtain the remote sensing image data, ground environmental data, manual evaluation data of the ground farmland construction quality in the corresponding image area, and the corresponding geographic coordinate data of the target area at different construction stages; the remote sensing image data includes multi-temporal high-resolution satellite images, radar images, and thermal infrared images; the ground environmental data includes infrastructure construction data, crop growth data, meteorological data, and disaster data;

[0012] Preprocess the remote sensing image data to obtain preliminary correction data; the preprocessing includes radiometric correction, geometric correction, image denoising, and image cropping;

[0013] Use GIS technology to map the ground environmental data, manual evaluation data, and the corresponding geographic coordinate data into the spatial coordinate system of the preliminary correction data for spatial registration, and use the method of linear interpolation for time alignment to obtain the final remote sensing environmental data.

[0014] Optionally, the high-standard farmland time-series collaborative sampling model specifically includes an input layer, a network layer, and an output layer;

[0015] The input layer is used to: input the remote sensing environmental data, set as (B, T, W, H, N), where B represents the batch size, T represents the time step, W and H respectively represent the width and height of the image, and N represents the number of channels of the image;

[0016] The network layer is used for: First, extracting multi-temporal remote sensing image features through the MobileNetV2 network, and outputting with a shape of (B, T, H', W', C1'), where H' and W' are the height and width of the feature map, and C1' is the number of feature channels; at the same time, extracting features of facilities such as ditches and roads through the U-Net network, and outputting with a shape of (B, T, H, W, C2); then, performing upsampling on the output of the MobileNetV2 network and adjusting it to (B, T, H, W, C1), and expanding the output channels of the U-Net network through 1x1 convolution to make C1 = C2; subsequently, splicing the two features in the channel dimension to obtain a feature with a shape of (B, T, H, W, C1 + C1), and compressing it to (B, T, H, W, C3) through 1x1 convolution; performing global average pooling on the compressed feature to obtain a feature vector with a shape of (B, T, C3); finally, using the meteorological time series data (B, T, C4) and the disaster time series data (B, T, C5) as auxiliary features, splicing them with the image features to obtain a fusion feature with a shape of (B, T, C3 + C4 + C5); inputting the feature into the LSTM network for time series modeling, outputting a feature representation with a shape of (B, T, hidden_size), and adding a Dropout layer after the LSTM to evaluate the uncertainty of the model;

[0017] The output layer is used for: processing through a fully connected layer and outputting the sampling inspection prediction result.

[0018] Optionally, the MobileNetV2 network specifically includes: a first convolutional layer, a bottleneck module, a second convolutional layer, and a global average pooling layer connected in sequence; where the bottleneck module includes 7 bottleneck layers; and each bottleneck layer is composed of a 1x1 convolution, a 3x3 depthwise separable convolution, and a 1x1 convolution.

[0019] Optionally, the U-Net network specifically includes: an encoder, a bottleneck layer, and a decoder connected in sequence; where the encoder is composed of multiple convolutional blocks, and each convolutional block includes two 3x3 convolutional layers, a ReLU activation function, and a 2x2 max pooling layer; the bottleneck layer includes two 3x3 convolutional layers and a ReLU activation function; the decoder is composed of multiple upsampling blocks, and each upsampling block includes a 2x2 transposed convolutional layer, a skip connection layer, two 3x3 convolutional layers, and a ReLU activation function.

[0020] The present invention also provides an intelligent sampling inspection system for farmland construction quality, including:

[0021] A data acquisition unit for acquiring remote sensing environmental data of the target area; the remote sensing environmental data includes multi-temporal remote sensing images and ground environmental data;

[0022] A model prediction unit for inputting the remote sensing environmental data into a high-standard farmland time-series collaborative sampling model to determine the sampling inspection prediction result; the prediction result includes a risk heat map and an uncertainty quantification map; the high-standard farmland time-series collaborative sampling model is constructed based on a variogram model, a time-series triggering mechanism, and Monte Carlo techniques for dynamically optimizing the sampling strategy;

[0023] A sampling inspection determination unit for performing uncertainty analysis based on the sampling inspection prediction result to generate a final sampling inspection area.

[0024] The present invention also provides an electronic device including a memory and a processor, where the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the intelligent sampling inspection method for farmland construction quality according to the above.

[0025] The present invention also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, it implements the intelligent sampling inspection method for farmland construction quality as described above.

[0026] According to the specific embodiments provided by the present invention, the following technical effects are disclosed:

[0027] The present invention discloses an intelligent sampling inspection method, system, device, and storage medium for farmland construction quality. The method includes obtaining remote sensing environmental data of a target area; the remote sensing environmental data includes multi-temporal remote sensing images and ground environmental data; inputting the remote sensing environmental data into a high-standard farmland time-series collaborative sampling model to determine the sampling inspection prediction result; the prediction result includes a risk heat map and an uncertainty quantification map; the high-standard farmland time-series collaborative sampling model is constructed based on a variogram model, a time-series triggering mechanism, and Monte Carlo techniques for dynamically optimizing the sampling strategy; performing uncertainty analysis based on the sampling inspection prediction result to generate a final sampling inspection area. The present invention can achieve the full-domain precise monitoring and dynamic optimization control of the high-standard farmland construction quality, providing technical support for the long-term guarantee of the farmland quality in the black soil area. Description of the Drawings

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0029] Figure 1 It is a schematic structural diagram of the high-standard farmland time-series collaborative sampling model in this embodiment;

[0030] Figure 2Schematic diagram of the MobileNetV2 network structure in this embodiment;

[0031] Figure 3 Schematic diagram of the U-Net network structure in this embodiment. Detailed implementation manners

[0032] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0033] The purpose of the present invention is to provide a method, system, device and storage medium for intelligent sampling inspection of farmland construction quality, which can realize the whole-region precise monitoring and dynamic optimization control of high-standard farmland construction quality, and provide technical support for the long-term guarantee of farmland quality in black soil areas.

[0034] To make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0035] As Figure 1 shown, the present invention provides a method for intelligent sampling inspection of farmland construction quality, including:

[0036] Step 100: Obtain remote sensing environmental data of the target area; the remote sensing environmental data includes multi-temporal remote sensing images and ground environmental data.

[0037] Step 200: Input the remote sensing environmental data into the high-standard farmland time-series collaborative sampling model to determine the sampling inspection prediction result; the prediction result includes a risk heat map and an uncertainty quantification map; the high-standard farmland time-series collaborative sampling model is constructed based on the variogram model, time-series trigger mechanism and Monte Carlo technique, and is used to dynamically optimize the sampling strategy.

[0038] Step 300: Perform uncertainty analysis based on the sampling inspection prediction result to generate the final sampling inspection area.

[0039] As a specific implementation manner, the specific operation processes of the above steps are as follows.

[0040] 1. Data acquisition and preprocessing

[0041] Obtain multi-temporal high-resolution satellite images, radar images, and thermal infrared images at different construction stages of high-standard farmland; obtain farmland construction data (such as the spatial distribution of infrastructure such as ditches, roads, and irrigation facilities); crop growth data (such as crop types, growth cycles, etc.); meteorological data (such as daily average temperature, rainfall, humidity, etc.); disaster data (such as the occurrence time and affected area of floods, droughts, etc.); obtain the manual evaluation data of the construction quality of high-standard farmland on the ground in the corresponding image area and the corresponding geographic coordinate data.

[0042] Preprocess the remote sensing images, including operations such as radiometric correction, geometric correction, image denoising, and image cropping; then perform spatial registration of the ground environmental data and the remote sensing images. Map the obtained ground environmental data and manual evaluation data into the spatial coordinate system of the preprocessed remote sensing images through the Geographic Information System (GIS); finally, perform temporal registration of the ground data and the remote sensing images. Use the method of linear interpolation to align the multi-temporal ground data with the multi-temporal remote sensing images in time.

[0043] 2. Import the multi-temporal remote sensing images and ground data processed in 1 into the high-standard farmland time-series collaborative sampling model constructed by the present invention, as Figure 1 shown, and introduce the Monte Carlo (Dropout) technique in model training. Randomly discard some neurons during the training process to evaluate the uncertainty of the model. The specific process includes:

[0044] Input layer: The model input includes multi-temporal remote sensing images and annotation data such as ditches and roads, and their shapes are (B, T, W, H, N) respectively, where B represents the batch size, T represents the time step, W and H represent the width and height of the image respectively, and N represents the number of channels of the image.

[0045] Network Structure: First, MobileNetV2 is used to extract features from multi-temporal remote sensing images, with an output shape of (B, T, H', W', C1'), where H' and W' are the height and width of the feature map, and C1' is the number of feature channels. At the same time, U-Net is used to extract features of facilities such as ditches and roads, with an output shape of (B, T, H, W, C2). Then, the output of MobileNetV2 is upsampled and adjusted to (B, T, H, W, C1), and the output channels of U-Net are expanded through 1x1 convolution to make the number of channels of the two the same (i.e., C1 = C2). Subsequently, the two features are concatenated in the channel dimension to obtain a feature with a shape of (B, T, H, W, C1 + C1), and it is compressed to (B, T, H, W, C3) through 1x1 convolution. Global average pooling (GAP) is performed on the compressed feature to obtain a feature vector with a shape of (B, T, C3). Further, meteorological time-series data (with a shape of (B, T, C4)) and disaster time-series data (with a shape of (B, T, C5)) are used as auxiliary features and concatenated with the image features to obtain a fused feature with a shape of (B, T, C3 + C4 + C5). This feature is input into LSTM for time-series modeling, with an output shape of (B, T, hidden_size), and a Dropout layer is added after LSTM to evaluate the uncertainty of the model.

[0046] Output Layer: After being processed by the fully connected layer, the model finally outputs the prediction results, including the risk heat map and the uncertainty quantification map. The risk heat map is used to identify potential risk areas in farmland construction, while the uncertainty quantification map evaluates the credibility of the model prediction through the Monte Carlo Dropout technique.

[0047] As Figure 2 shown, the MobileNetV2 network includes: The initial convolutional layer Conv2d is a 3×3 convolutional layer with stride = 2, which is used to initially extract features; The bottleneck layer (BottleneckBlocks) consists of 1x1 convolution, 3x3 depthwise separable convolution (Depthwise Conv), and 1x1 convolution; Global average pooling (GlobalAverage Pooling, GAP) is used to perform global average pooling on the feature map of each channel to compress the spatial dimension; The last Conv layer is used to adjust the number of output channels and output the result.

[0048] As Figure 3As shown in the figure, the U-Net network includes: The encoder consists of multiple convolutional blocks, each convolutional block contains two 3x3 convolutional layers (Conv2d), ReLU activation function, and 2x2 max pooling (MaxPooling); The bottleneck layer is located between the encoder and the decoder, and contains two 3x3 convolutional layers (Conv2d) and ReLU activation function; The decoder consists of multiple upsampling blocks, each upsampling block contains 2x2 up-convolutional layer (UpConv), skip connection (Copy and Crop), two 3x3 convolutional layers (Conv2d) and ReLU activation function; The last Conv is used to map the number of channels to the number of target classes.

[0049] 3. Prediction value space correction. The variogram model is constructed by using Kriging Interpolation method to fully exploit the autocorrelation and distribution law of spatial data, and the optimal unbiased estimation is carried out for unknown points. The formula for constructing the variogram is as follows:

[0050]

[0051] Among them, c0 represents the micro-scale variation or measurement error, c represents the heterogeneity caused by spatial autocorrelation, and a represents the range of spatial autocorrelation. By analyzing the spatial structure characteristics of the measured data, this method generates a continuous spatial distribution map and quantifies the uncertainty of the interpolation results, effectively solving the prediction deviation problem caused by complex terrain or uneven data distribution in the time-series collaborative sampling model of high-standard farmland, and significantly improving the accuracy and reliability of the prediction results.

[0052] 4. Embedding the time-series trigger mechanism. Design a time-series rule library, use the rule engine to dynamically trigger the adjustment of the sampling strategy, and embed the time-series trigger mechanism (such as focusing on spot-checking areas with complex terrain during the construction period; focusing on spot-checking areas with abnormal crop growth during the management period; focusing on spot-checking fertilization and irrigation areas during certain phenological periods, etc.) into the time-series collaborative sampling model. This mechanism automatically adjusts the sampling frequency and spatial center-of-gravity distribution of the model input data by real-time monitoring the changes of time nodes and farmland construction stages, enabling the model to adapt to the needs of different stages. The input data optimized by the time-series trigger mechanism directly affects the prediction results of the model (such as risk heat map, uncertainty quantification map), and the prediction result display module takes these results as input to generate visual results and the final sampling area.

[0053] 5. Prediction result display. After the model completes the prediction, it directly outputs the risk heat map and the uncertainty quantification map. At the same time, combining the prediction results with the uncertainty analysis, it automatically generates the final sampling area. Through the integrated interactive visualization tool, users can dynamically adjust the threshold ratio of uncertainty quantification according to the needs of different periods, and flexibly optimize the division of the sampling area.

[0054] Therefore, the present solution has the following beneficial effects:

[0055] By integrating multi-temporal remote sensing images and ground environmental data, combining the time-series trigger mechanism and Monte Carlo Dropout technology, dynamically optimizing the sampling strategy, and automatically generating sampling areas based on the risk heat map and uncertainty quantification map, this method realizes intelligent and precise sampling inspection of a large area of high-standard farmland.

[0056] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same and similar parts among the various embodiments, reference can be made to each other.

[0057] Specific examples are used in this article to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. An intelligent sampling inspection method for the quality of farmland construction, characterized in that, Including: Obtain the remote sensing environmental data of the target area; the remote sensing environmental data includes multi-temporal remote sensing images and ground environmental data; Input the remote sensing environmental data into the high-standard farmland time-series collaborative sampling model to determine the sampling inspection prediction result; the prediction result includes a risk heat map and an uncertainty quantification map; the high-standard farmland time-series collaborative sampling model is constructed based on a variogram model, a time-series trigger mechanism, and Monte Carlo techniques and is used to dynamically optimize the sampling strategy; Conduct uncertainty analysis based on the sampling inspection prediction result to generate the final sampling inspection area.

2. The intelligent sampling inspection method for farmland construction quality according to claim 1, wherein The specific process of obtaining the remote sensing environmental data of the target area includes: Obtain the remote sensing image data, ground environmental data, manual evaluation data of the ground farmland construction quality in the corresponding image area, and the corresponding geographic coordinate data of the target area at different construction stages; the remote sensing image data includes multi-temporal high-resolution satellite images, radar images, and thermal infrared images; the ground environmental data includes infrastructure construction data, crop growth data, meteorological data, and disaster data; Preprocess the remote sensing image data to obtain preliminary calibration data; the preprocessing includes radiometric calibration, geometric correction, image denoising, and image cropping; Use GIS technology to map the ground environmental data, manual evaluation data, and the corresponding geographic coordinate data into the spatial coordinate system of the preliminary calibration data for spatial registration, and use the method of linear interpolation for time alignment to obtain the final remote sensing environmental data.

3. The intelligent sampling inspection method for farmland construction quality according to claim 1, wherein The high-standard farmland time-series collaborative sampling model specifically includes: an input layer, a network layer, and an output layer; The input layer is used to: input the remote sensing environmental data, set as (B, T, W, H, N), where B represents the batch size, T represents the time step, W and H respectively represent the width and height of the image, and N represents the number of channels of the image; The network layer is used for: First, extracting multi-temporal remote sensing image features through the MobileNetV2 network, and outputting with a shape of (B, T, H', W', C1'), where H' and W' are the height and width of the feature map, and C1' is the number of feature channels; at the same time, extracting features of facilities such as ditches and roads through the U-Net network, and outputting with a shape of (B, T, H, W, C2); then, performing upsampling on the output of the MobileNetV2 network to adjust it to (B, T, H, W, C1), and expanding the output channels of the U-Net network through 1x1 convolution to make C1 = C2; subsequently, concatenating the two features in the channel dimension to obtain a feature with a shape of (B, T, H, W, C1 + C1), and compressing it to (B, T, H, W, C3) through 1x1 convolution; performing global average pooling on the compressed feature to obtain a feature vector with a shape of (B, T, C3); finally, taking the meteorological time series data (B, T, C4) and the disaster time series data (B, T, C5) as auxiliary features, and concatenating them with the image features to obtain a fusion feature with a shape of (B, T, C3 + C4 + C5); inputting the feature into the LSTM network for time series modeling, outputting a feature representation with a shape of (B, T, hidden_size), and adding a Dropout layer after the LSTM to evaluate the uncertainty of the model; The output layer is used for: Processing through a fully connected layer and outputting the sampling inspection prediction result.

4. The intelligent sampling inspection method for farmland construction quality according to claim 3, characterized in that, The MobileNetV2 network specifically includes: A first convolutional layer, a bottleneck module, a second convolutional layer, and a global average pooling layer connected in sequence; where the bottleneck module includes 7 bottleneck layers; and each bottleneck layer consists of a 1x1 convolution, a 3x3 depthwise separable convolution, and a 1x1 convolution.

5. The intelligent sampling inspection method for farmland construction quality according to claim 3, characterized in that The U-Net network specifically includes: An encoder, a bottleneck layer, and a decoder connected in sequence; where the encoder consists of multiple convolutional blocks, and each convolutional block includes two 3x3 convolutional layers, a ReLU activation function, and a 2x2 max pooling layer; the bottleneck layer includes two 3x3 convolutional layers and a ReLU activation function; the decoder consists of multiple upsampling blocks, and each upsampling block includes a 2x2 transposed convolutional layer, a skip connection layer, two 3x3 convolutional layers, and a ReLU activation function.

6. An intelligent sampling inspection system for the quality of farmland construction, characterized in that, Including: A data acquisition unit for acquiring remote sensing environmental data of the target area; the remote sensing environmental data includes multi-temporal remote sensing images and ground environmental data; A model prediction unit for inputting the remote sensing environmental data into the high-standard farmland time series collaborative sampling model to determine the sampling inspection prediction result; the prediction result includes a risk heat map and an uncertainty quantification map; the high-standard farmland time series collaborative sampling model is constructed based on a variogram model, a time series trigger mechanism, and Monte Carlo techniques for dynamically optimizing the sampling strategy; A sampling inspection determination unit for performing uncertainty analysis based on the sampling inspection prediction result and generating the final sampling inspection area.

7. An electronic device, characterized in that, It includes a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the intelligent sampling inspection method for the quality of farmland construction according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, It stores a computer program, and when the computer program is executed by a processor, it realizes the intelligent sampling inspection method for the quality of farmland construction according to any one of claims 1-5.

Images