Crop irrigation demand prediction method based on intelligent integrated prefabricated pump station

Abstract

The invention relates to a crop irrigation demand prediction method based on an intelligent integrated prefabricated pump station, and belongs to the field of agricultural irrigation. The method comprises the following steps: acquiring crop irrigation demand data, and preprocessing the data to obtain a crop irrigation demand data set; crop images in the crop irrigation demand data set are processed by a multi-scale visual feature fusion module to obtain image color features, image edge features, image shape features and high-dimensional comprehensive features, and multi-scale visual fusion features are obtained through feature fusion; environment features in the crop irrigation demand data set pass through an environment feature representation and fusion module to obtain environment feature representation vectors, and then the multi-scale visual fusion features and the environment feature representation vectors are spliced to obtain multi-scale fusion features; the multi-scale fusion features are decided through a decision module, and a crop irrigation strategy prediction result is obtained. The crop irrigation efficiency can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of agricultural irrigation, and in particular relates to a crop irrigation demand prediction method based on an intelligent integrated prefabricated pump station. Background Art

[0002] An integrated prefabricated pump station is a pump station that integrates the main components such as wellbore, pump, pipeline, control system and ventilation system into one in the factory, and is pre-installed and tested before leaving the factory. It is suitable for water supply, drainage and waterlogging prevention projects. The pump station is partially or completely buried underground, with inspection holes and control boxes on the ground. It can also be equipped with a landscape management room placed on the ground; the equipment has low noise and occupies a small area; the integrated prefabricated pump station equipment is easy to transport and hoist, and the construction period is much shorter than that of traditional pump stations. Intelligent integrated prefabricated pump station is to achieve more intelligent irrigation on the basis of integrated prefabricated pump station. With the modernization and intelligent development of agricultural production, traditional irrigation methods are gradually unable to meet the needs of high efficiency and water conservation. In recent years, computer vision technology has been increasingly widely used in agriculture. It can provide accurate irrigation demand data through real-time monitoring of crop growth conditions. At present, the irrigation demand of crops mainly depends on manual monitoring and empirical judgment of factors such as soil moisture, meteorological conditions and crop growth stage. However, this method is not only inefficient, but also easily affected by human factors, resulting in uneven irrigation and waste of water resources. Therefore, a new irrigation demand prediction method is urgently needed to achieve accurate and intelligent crop irrigation management. The emergence of intelligent integrated prefabricated pumping stations provides a more flexible and efficient solution for irrigation systems. By combining computer vision technology, it is possible to automatically monitor the growth status of crops, quickly obtain crop growth information, and intelligently adjust irrigation strategies based on real-time data. This can not only improve the efficiency of irrigation, but also effectively save water resources and promote the development of sustainable agriculture.

[0003] Intelligent integrated prefabricated pump station technology has great practical value in the construction and management of water conservancy projects, but it is rarely used in agricultural irrigation and drainage projects. This project aims to explore the application prospects of intelligent integrated prefabricated pump stations in agricultural irrigation and drainage projects and actively promote them. It provides a crop irrigation demand prediction method based on intelligent integrated prefabricated pump stations, uses advanced image processing methods to extract crop visual features and automatically detect crop growth status in real time, and integrates visual features with environmental features detected by sensors to train decision modules so that they can give accurate and reasonable irrigation plans. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a crop irrigation demand prediction method based on an intelligent integrated prefabricated pump station.

[0005] In order to achieve the above object, the present invention is implemented through the following technical solutions: The present invention provides a method for predicting crop irrigation demand based on an intelligent integrated prefabricated pump station, comprising the following steps: S1. Acquire crop irrigation demand data, pre-process the data, and obtain a crop irrigation demand data set; S2. The crop images in the crop irrigation demand dataset are processed by a multi-scale visual feature fusion module to obtain image color features, image edge features, image shape features and high-dimensional comprehensive features, and multi-scale visual fusion features are obtained by feature fusion; the multi-scale visual feature fusion module includes a feature extraction module and a feature fusion module; the feature extraction module includes a color feature extraction unit, an edge feature extraction unit, a shape feature extraction unit and a high-dimensional comprehensive feature extraction unit; S3. The environmental features in the crop irrigation demand data set are subjected to an environmental feature characterization and fusion module to obtain an environmental feature characterization vector, and then the multi-scale visual fusion feature and the environmental feature characterization vector are spliced ​​to obtain a multi-scale fusion feature; S4. The multi-scale fusion features are used for decision making by the decision module to obtain the prediction result of the crop irrigation strategy.

[0006] Further, step S1 includes: The crop irrigation demand data includes crop images at different growth periods, environmental characteristics, and irrigation strategies corresponding to each image; the different growth periods include germination period, seedling period, growth period, flowering period, fruiting period, and maturity period; the environmental characteristics are environmental characteristics of the 7 days before the picture is taken; the irrigation strategies corresponding to each image include no irrigation: 0, slight irrigation: 1, medium irrigation: 2, and heavy irrigation: 3; the growth period of each crop image is manually labeled; Preprocess the labeled crop images to obtain a crop irrigation demand dataset that can be recognized and processed by computers. ; The preprocessing operations include image denoising, image enhancement and image scaling.

[0007] Furthermore, in step S2, color feature extraction: the image in the crop irrigation demand data set Input into the color feature extraction unit, extract the color features of the input image through the color histogram, and obtain the image color features , the formula is as follows: , , , in, Indicated in The number of pixels in an interval, Representing images In Location The pixel color at The color space intervals, is the Kronecker Delta function, used to check Is it , when the pixel The color value falls within the range When , the value is 1, otherwise the value is 0; represents the normalized color feature histogram vector, represents the total number of pixels in the image, Represents the color characteristics of the image.

[0008] Furthermore, in step S2, edge feature extraction: the image in the crop irrigation demand data set Input it into the edge feature extraction unit, and use the Canny edge detection method to extract the edge features of the input image to obtain the image edge features Specifically: First, the image The grayscale image is converted to a grayscale image, and the grayscale image is smoothed by a Gaussian filter to obtain a filtered image , and then calculate the filtered image through the Sobel operator The gradient magnitude in the x and y axis directions, and then the gradient strength and direction of each pixel are obtained; the minimum value is used and maximum value These two thresholds classify the edges. It is marked as a strong edge. It is marked as a weak edge. Then mark it as non-edge, connect weak edges with strong edges, retain those weak edges connected to strong edges, and finally get the image edge features ; The formula is as follows: , , , , , , , in, Indicates the position in edge feature extraction The result after filtering is represents a Gaussian filter, k and l represent the window size in the x and y axis directions, respectively, which determines the size of the filter. Represent the gradient amplitude in the x and y axis directions respectively, and Respectively represent the filters of the Sobel operator in the x and y directions, represents the gradient strength, represents the gradient direction, Representing images In Location Where value, Represents the length and width of the image respectively.

[0009] Furthermore, in step S2, shape feature extraction: the image in the crop irrigation demand data set Input into the shape feature extraction unit, extract the shape features of the input image through the local binary pattern, and obtain the image shape features , specifically: for the image Each pixel in , define a window size of 3×3 and calculate the LBP value of each pixel. The formula is as follows: , in, Representing images The selected pixel in represents the neighborhood pixels around the central pixel, Indicates the use for judgment and The indicator function of the size value between hour, The value is 1, otherwise it is 0; assign the LBP value of each pixel to the corresponding position to obtain the LBP image , calculate the image The histogram of is expressed as follows: , in, Representing images The histogram of Indicates the use for judging Is it equal to The indicator function is represented by hour, The value is 1, otherwise it is 0; the histogram is normalized to obtain the image shape characteristics , the formula is as follows: , , in, represents the total number of pixels in the histogram, It represents the frequency of occurrence of the i-th LBP mode, and there are P LBP modes in total.

[0010] Furthermore, in step S2, the high-dimensional comprehensive feature extraction unit is a ResNet34 neural network; the image in the crop irrigation demand data set Input into the high-dimensional comprehensive feature extraction unit to obtain high-dimensional comprehensive features , the formula is as follows: , in, ResNet34 neural network The weight matrix of the layer, ResNet34 neural network The output of the layer, ResNet34 neural network The bias vector for the layer.

[0011] Furthermore, the image color feature , image edge features , image shape features And high-dimensional comprehensive features Input into the feature fusion module for feature fusion to obtain multi-scale visual fusion features , the formula is as follows: , , , in, Representation characteristics High-dimensional comprehensive features extracted with ResNet34 neural network The cos similarity between Representation characteristics The weight of Indicates film removal operation. Represents multi-scale visual fusion features.

[0012] Further, step S3 includes: Environmental characteristics include temperature characteristics, humidity characteristics and soil moisture characteristics; the temperature characteristics include the maximum temperature , minimum temperature , average temperature , the maximum temperature of each day , minimum temperature , average temperature Mapped three-dimensional temperature characteristics The temperature characteristics of a day are obtained by comparing and calculating the 24 collected data values ​​from the temperature sensor starting from 0:00 every day and collecting data every 1 hour; similarly, the three-dimensional humidity characteristics and three-dimensional soil humidity characteristics of each day are obtained. ; Construct daily temperature characteristics, humidity characteristics, and soil moisture characteristics into a three-dimensional vector , where i∈{1,2,3,4,5,6,7} represents the i-th day, represents the environmental characteristics of the i-th day when the photo was taken, represents the temperature characteristics of the i-th day when the photo was taken, shows the humidity characteristics of the i-th day when the photo was taken, represents the soil moisture characteristics on the i-th day when the photo was taken; The three-dimensional vector The environmental characteristics of the first 7 days are represented by the long short-term memory neural network LSTM model to obtain the environmental characteristics representation vector ; Multi-scale visual fusion features and environmental characterization vector Splice to obtain multi-scale fusion features , the formula is as follows: .

[0013] Further, step S4 includes: The decision-making modules include vector machines (SVM), random forests (RF), k-nearest neighbor algorithms (KNN), logistic regression (LR) and multi-layer perceptrons (MLP); The multi-scale fusion feature After the decision-making module, the prediction results of crop irrigation strategy are obtained , the formula is as follows: in, represents the prediction result of crop irrigation strategy, represents the irrigation strategy category, represents the probability that the i-th classifier predicts that it belongs to the k-th irrigation strategy category, is the average predicted probability of irrigation strategy category k, and argmax means taking the maximum value.

[0014] The advantages of the present invention are: The present invention simultaneously considers the growth status (whether abnormal), growth state (stage) and environment of crops to achieve accurate prediction of the amount of irrigation water required for them; in the visual feature extraction module, a similar weight distribution mechanism is used to better extract the multi-system features of the image; the visual features extracted from the crop image are spliced ​​with the environmental features of the photo 7 days before the photo was taken obtained by LSTM characterization to obtain a multi-scale fusion feature that integrates the visual and environmental features; late fusion: a multi-decision maker fusion method is used in the prediction stage to better utilize the multi-scale fusion features to obtain the final irrigation prediction result; the prediction effect of crop irrigation is improved, and the efficiency of crop irrigation can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0016] Figure 1 is a flow chart of the method of the present invention; Figure 2 The following are data examples corresponding to various irrigation strategies of the method of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Example 1 In this embodiment, Figure 1 As shown, the present invention provides a crop irrigation demand prediction method based on an intelligent integrated prefabricated pump station, and the specific steps include: S1. Acquire crop irrigation demand data, pre-process the data, and obtain a crop irrigation demand data set; Specifically, the crop irrigation demand data includes crop images at different growth periods, environmental characteristics, and irrigation strategies corresponding to each image; the different growth periods include germination period, seedling period, growth period, flowering period, fruiting period, and maturity period; the environmental characteristics are environmental characteristics of the 7 days before the picture is taken; the irrigation strategies corresponding to each image include no irrigation: 0, slight irrigation: 1, medium irrigation: 2, and heavy irrigation: 3; the growth period of each crop image is manually labeled; Preprocess the labeled crop images to obtain a crop irrigation demand dataset that can be recognized and processed by computers. ; The preprocessing operations include image denoising, image enhancement and image scaling.

[0019] S2. The crop images in the crop irrigation demand dataset are processed by a multi-scale visual feature fusion module to obtain image color features, image edge features, image shape features and high-dimensional comprehensive features, and multi-scale visual fusion features are obtained by feature fusion; the multi-scale visual feature fusion module includes a feature extraction module and a feature fusion module; the feature extraction module includes a color feature extraction unit, an edge feature extraction unit, a shape feature extraction unit and a high-dimensional comprehensive feature extraction unit; Specifically, color feature extraction: the crop irrigation demand dataset contains images Input into the color feature extraction unit, extract the color features of the input image through the color histogram, and obtain the image color features , the formula is as follows: , , , in, Indicated in The number of pixels in an interval, Representing images In Location The pixel color at The color space intervals, is the Kronecker Delta function, used to check Is it , when the pixel The color value falls within the range When , the value is 1, otherwise the value is 0; represents the normalized color feature histogram vector, represents the total number of pixels in the image, Represents the color characteristics of the image.

[0020] Specifically, edge feature extraction: the crop irrigation demand dataset contains images Input it into the edge feature extraction unit, and use the Canny edge detection method to extract the edge features of the input image to obtain the image edge features Specifically: First, the image The grayscale image is converted to a grayscale image, and the grayscale image is smoothed by a Gaussian filter to obtain a filtered image , and then calculate the filtered image through the Sobel operator The gradient magnitude in the x and y axis directions, and then the gradient strength and direction of each pixel are obtained; the minimum value is used and maximum value These two thresholds classify the edges. It is marked as a strong edge. It is marked as a weak edge. Then mark it as non-edge, connect weak edges with strong edges, retain those weak edges connected to strong edges, and finally get the image edge features ; The formula is as follows: , , , , , , , in, Indicates the position in edge feature extraction The result after filtering is represents a Gaussian filter, k and l represent the window size in the x and y axis directions, respectively, which determines the size of the filter. Represent the gradient amplitude in the x and y axis directions respectively, and Respectively represent the filters of the Sobel operator in the x and y directions, represents the gradient strength, represents the gradient direction, Representing images In Location Where value, Represents the length and width of the image respectively.

[0021] Specifically, shape feature extraction: the crop irrigation demand dataset contains images Input into the shape feature extraction unit, extract the shape features of the input image through the local binary pattern, and obtain the image shape features , specifically: for the image Each pixel in , define a window size of 3×3 and calculate the LBP value of each pixel. The formula is as follows: , in, Representing images The selected pixel in represents the neighborhood pixels around the central pixel, Indicates the use for judgment and The indicator function of the size value between hour, The value is 1, otherwise it is 0; assign the LBP value of each pixel to the corresponding position to obtain the LBP image , calculate the image The histogram of is expressed as follows: , in, Representing images The histogram of Indicates the use for judging Is it equal to The indicator function is represented by hour, The value is 1, otherwise it is 0; the histogram is normalized to obtain the image shape characteristics , the formula is as follows: , , in, represents the total number of pixels in the histogram, It represents the frequency of occurrence of the i-th LBP mode, and there are P LBP modes in total.

[0022] Specifically, the high-dimensional comprehensive feature extraction unit is a ResNet34 neural network; High-dimensional comprehensive feature extraction: images in the crop irrigation demand dataset Input into the high-dimensional comprehensive feature extraction unit to obtain high-dimensional comprehensive features , the formula is as follows: , in, ResNet34 neural network The weight matrix of the layer, ResNet34 neural network The output of the layer, ResNet34 neural network The bias vector for the layer.

[0023] Specifically, the image color feature , image edge features , image shape features And high-dimensional comprehensive features Input into the feature fusion module for feature fusion to obtain multi-scale visual fusion features , the formula is as follows: , , , in, Representation characteristics High-dimensional comprehensive features extracted with ResNet34 neural network The cos similarity between Representation characteristics The weight of Indicates film removal operation. Represents multi-scale visual fusion features.

[0024] S3. The environmental features in the crop irrigation demand data set are subjected to an environmental feature characterization and fusion module to obtain an environmental feature characterization vector, and then the multi-scale visual fusion feature and the environmental feature characterization vector are spliced ​​to obtain a multi-scale fusion feature; Specifically, the environmental characteristics include temperature characteristics, humidity characteristics and soil moisture characteristics; the temperature characteristics include the maximum temperature , minimum temperature , average temperature , the maximum temperature of each day , minimum temperature , average temperature Mapped three-dimensional temperature characteristics The temperature characteristics of a day are obtained by comparing and calculating the 24 collected data values ​​from the temperature sensor starting from 0:00 every day and collecting data every 1 hour; similarly, the three-dimensional humidity characteristics and three-dimensional soil humidity characteristics of each day are obtained. ; Construct daily temperature characteristics, humidity characteristics, and soil moisture characteristics into a three-dimensional vector , where i∈{1,2,3,4,5,6,7} represents the i-th day, represents the environmental characteristics of the i-th day when the photo was taken, represents the temperature characteristics of the i-th day when the photo was taken, shows the humidity characteristics of the i-th day when the photo was taken, represents the soil moisture characteristics on the i-th day when the photo was taken; The three-dimensional vector The environmental characteristics of the first 7 days are represented by the long short-term memory neural network LSTM model to obtain the environmental characteristics representation vector ; Multi-scale visual fusion features and environmental characterization vector Splice to obtain multi-scale fusion features , the formula is as follows: .

[0025] S4. The multi-scale fusion features are used for decision making by the decision module to obtain the prediction result of the crop irrigation strategy.

[0026] Specifically, the decision module includes vector machine SVM, random forest RF, k nearest neighbor algorithm KNN, logistic regression LR and multi-layer perceptron MLP; The multi-scale fusion feature After the decision-making module, the prediction results of crop irrigation strategy are obtained , the formula is as follows: in, represents the prediction result of crop irrigation strategy, represents the irrigation strategy category, represents the probability that the i-th classifier predicts that it belongs to the k-th irrigation strategy category, is the average predicted probability of irrigation strategy category k, and argmax means taking the maximum value.

[0027] Example 2 In this embodiment, crop growth status data is collected by manual photography, environmental features are collected by sensors, and the collected image data are labeled with four labels: [no irrigation: 0], [light irrigation: 1], [medium irrigation: 2], and [large irrigation: 3]. The shooting time is recorded for each image, and LSTM is used to characterize the environmental features of the 7 days before the shooting time. Give the image, such as Figure 2 As shown in the figure, during the environmental feature characterization process, sensors are used to obtain environmental features (including temperature, humidity, and soil moisture) at intervals of one hour and stored in the database for backup. The acquisition time is recorded when each crop image is acquired, and the environmental features of the photo taken 7 days before are found. , which is used as the input of LSTM to obtain a representation vector containing the environmental features of the previous 7 days when the image was taken , Contains temperature characteristics for the previous 7 days , humidity characteristics , soil moisture characteristics The collected crop images are passed through the irrigation analysis module of the intelligent integrated prefabricated pump station. By analyzing the growth status and environmental characteristics of the crops, the crop irrigation strategy can be predicted. Then, the irrigation control module in the intelligent integrated prefabricated pump station irrigates the crops according to the predicted results of the irrigation strategy.

[0028] To verify the effectiveness of the present invention, the data set is divided into a training set and a test set at a ratio of 3:1. In the decision module, different types of decision makers are trained using the multi-scale fusion features of the fused visual features and environmental features extracted from the training set, and the trained decision makers are used to make decisions. Table 1 shows the performance of the present method and several individual decision makers on the test set.

[0029] Table 1 Comparison of experimental data between the proposed method and a single decision maker Table 1 shows the accuracy, recall rate and F1 value of irrigation strategies obtained by selecting different decision makers in the decision module. From the results in Table 1, it can be seen that different decision makers trained with fusion features can achieve good performance, and the comprehensive performance of the present invention is better than that of a single decision maker. This illustrates the effectiveness of the feature extraction method and decision fusion method proposed in the present invention, and proves the effectiveness of the crop irrigation demand prediction method based on the intelligent integrated prefabricated pump station proposed in the present invention.

[0030] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting crop irrigation demand based on an intelligent integrated prefabricated pump station, characterized in that: The following steps are involved: S1. Acquire crop irrigation demand data, pre-process the data, and obtain a crop irrigation demand data set; S2. The crop images in the crop irrigation demand dataset are processed by a multi-scale visual feature fusion module to obtain image color features, image edge features, image shape features and high-dimensional comprehensive features, and multi-scale visual fusion features are obtained by feature fusion; the multi-scale visual feature fusion module includes a feature extraction module and a feature fusion module; The feature extraction module includes a color feature extraction unit, an edge feature extraction unit, a shape feature extraction unit and a high-dimensional comprehensive feature extraction unit; S3. The environmental features in the crop irrigation demand data set are subjected to an environmental feature characterization and fusion module to obtain an environmental feature characterization vector, and then the multi-scale visual fusion feature and the environmental feature characterization vector are spliced ​​to obtain a multi-scale fusion feature; S4. The multi-scale fusion features are used for decision making by the decision module to obtain the prediction result of the crop irrigation strategy.

2. The method for predicting crop irrigation demand based on intelligent integrated prefabricated pump station according to claim 1 is characterized in that: Step S1 includes: The crop irrigation demand data includes crop images at different growth periods, environmental characteristics, and irrigation strategies corresponding to each image; the different growth periods include germination period, seedling period, growth period, flowering period, fruiting period, and maturity period; the environmental characteristics are environmental characteristics of the 7 days before the picture is taken; the irrigation strategies corresponding to each image include no irrigation: 0, slight irrigation: 1, medium irrigation: 2, and heavy irrigation: 3; the growth period of each crop image is manually labeled; Preprocess the labeled crop images to obtain a crop irrigation demand dataset that can be recognized and processed by computers. ; The preprocessing operations include image denoising, image enhancement and image scaling.

3. The method for predicting crop irrigation demand based on intelligent integrated prefabricated pump station according to claim 2 is characterized in that: The color feature extraction in step S2 is specifically as follows: The crop irrigation demand dataset contains images Input into the color feature extraction unit, extract the color features of the input image through the color histogram, and obtain the image color features , the formula is as follows: ， ， ， in, Indicated in The number of pixels in an interval, Representing images In Location The pixel color at The color space intervals, is the Kronecker Delta function, which is used to check Is it , when the pixel The color value falls within the range When , the value is 1, otherwise the value is 0; represents the normalized color feature histogram vector, represents the total number of pixels in the image, Represents the color characteristics of the image.

4. The method for predicting crop irrigation demand based on intelligent integrated prefabricated pump station according to claim 3 is characterized in that: The edge feature extraction in step S2 is specifically as follows: The crop irrigation demand dataset contains images Input it into the edge feature extraction unit, and use the Canny edge detection method to extract the edge features of the input image to obtain the image edge features Specifically: First, the image The grayscale image is converted to a grayscale image, and the grayscale image is smoothed by a Gaussian filter to obtain a filtered image , and then calculate the filtered image through the Sobel operator The gradient magnitude in the x and y axis directions, and then the gradient strength and direction of each pixel; Use minimum value and maximum value These two thresholds classify the edges. It is marked as a strong edge. It is marked as a weak edge. Then mark it as non-edge, connect weak edges with strong edges, retain those weak edges connected to strong edges, and finally get the image edge features ; The formula is as follows: ， ， ， ， ， ， ， in, Indicates the position in edge feature extraction The result after filtering is represents a Gaussian filter, k and l represent the window size in the x and y axis directions, respectively, which determines the size of the filter. Represent the gradient amplitude in the x and y axis directions respectively, and Respectively represent the filters of the Sobel operator in the x and y directions, represents the gradient strength, represents the gradient direction, Representing images In Location Where value, Represents the length and width of the image respectively.

5. The method for predicting crop irrigation demand based on intelligent integrated prefabricated pump station according to claim 4 is characterized in that: The shape feature extraction in step S2 is specifically as follows: The crop irrigation demand dataset contains images Input into the shape feature extraction unit, extract the shape features of the input image through the local binary pattern, and obtain the image shape features , specifically: for the image Each pixel in , define a window size of 3×3 and calculate the LBP value of each pixel. The formula is as follows: ， in, Representing images The selected pixel in represents the neighborhood pixels around the central pixel, Indicates the use for judgment and The indicator function of the size value between hour, The value is 1, otherwise it is 0; assign the LBP value of each pixel to the corresponding position to obtain the LBP image , calculate the image The histogram of is expressed as follows: ， in, Representing images The histogram of Indicates the use for judging Is it equal to The indicator function is represented by hour, The value is 1, otherwise it is 0; the histogram is normalized to obtain the image shape characteristics , the formula is as follows: ， ， in, represents the total number of pixels in the histogram, It represents the frequency of occurrence of the i-th LBP mode, and there are P LBP modes in total.

6. The method for predicting crop irrigation demand based on intelligent integrated prefabricated pump station according to claim 5, characterized in that: The high-dimensional comprehensive feature extraction in step S2 is specifically as follows: The high-dimensional comprehensive feature extraction unit is a ResNet34 neural network; The crop irrigation demand dataset contains images Input into the high-dimensional comprehensive feature extraction unit to obtain high-dimensional comprehensive features , the formula is as follows: ， in, ResNet34 neural network The weight matrix of the layer, ResNet34 neural network The output of the layer, ResNet34 neural network The bias vector for the layer.

7. The method for predicting crop irrigation demand based on intelligent integrated prefabricated pump station according to claim 6, characterized in that: In step S2, the multi-scale visual fusion feature obtained by feature fusion is specifically: The image color feature , image edge features , image shape features And high-dimensional comprehensive features Input into the feature fusion module for feature fusion to obtain multi-scale visual fusion features , the formula is as follows: ， ， ， in, Representation characteristics High-dimensional comprehensive features extracted with ResNet34 neural network The cos similarity between Representation characteristics The weight of Indicates the film taking operation. Represents multi-scale visual fusion features.

8. The method for predicting crop irrigation demand based on intelligent integrated prefabricated pump station according to claim 7, characterized in that: Step S3 includes: Environmental characteristics include temperature characteristics, humidity characteristics and soil moisture characteristics; the temperature characteristics include the maximum temperature , minimum temperature , average temperature , the maximum temperature of each day , minimum temperature , average temperature Mapped three-dimensional temperature characteristics The temperature characteristics of a day are obtained by comparing and calculating the 24 collected data values ​​from the temperature sensor starting from 0:00 every day and collecting data every 1 hour; similarly, the three-dimensional humidity characteristics and three-dimensional soil humidity characteristics of each day are obtained. ; Construct daily temperature characteristics, humidity characteristics, and soil moisture characteristics into a three-dimensional vector , where i∈{1,2,3,4,5,6,7} represents the i-th day, represents the environmental characteristics of the i-th day when the photo was taken, represents the temperature characteristics of the i-th day when the photo was taken, shows the humidity characteristics of the i-th day when the photo was taken, represents the soil moisture characteristics on the i-th day when the photo was taken; The three-dimensional vector The environmental characteristics of the first 7 days are represented by the long short-term memory neural network LSTM model to obtain the environmental characteristics representation vector ; Multi-scale visual fusion features and environmental characterization vector Splice to obtain multi-scale fusion features , the formula is as follows: 。 9. The method for predicting crop irrigation demand based on intelligent integrated prefabricated pump station according to claim 8, characterized in that: Step S4 includes: The decision-making modules include vector machines (SVM), random forests (RF), k-nearest neighbor algorithms (KNN), logistic regression (LR) and multi-layer perceptrons (MLP); The multi-scale fusion feature After the decision-making module, the prediction results of crop irrigation strategy are obtained , the formula is as follows: in, represents the prediction result of crop irrigation strategy, represents the irrigation strategy category, represents the probability that the i-th classifier predicts that it belongs to the k-th irrigation strategy category, is the average predicted probability of irrigation strategy category k, and argmax means taking the maximum value.

Images